paper-with-me

Object Detection 벤치마크

Object Detection on COCO minival

1100개 결과 · ⬇ CSV · JSON

box AP

33.2 41.4 49.6 57.8 66 2015-12 2026-09 Cascade Mask R-CNN (ResNet-50) — 46.3 (2015-12-10) GFL (ResNet-50) — 44.5 (2015-12-10) ATSS (ResNet-50) — 43.5 (2015-12-10) Cascade Mask R-CNN (ResNet-50) — 46.3 (2015-12-10) GFL (ResNet-50) — 44.5 (2015-12-10) ATSS (ResNet-50) — 43.5 (2015-12-10) Cascade Mask R-CNN (ResNet-50) — 46.3 (2015-12-10) GFL (ResNet-50) — 44.5 (2015-12-10) ATSS (ResNet-50) — 43.5 (2015-12-10) Cascade Mask R-CNN (ResNet-50) — 46.3 (2015-12-10) GFL (ResNet-50) — 44.5 (2015-12-10) ATSS (ResNet-50) — 43.5 (2015-12-10) Cascade Mask R-CNN (ResNet-50) — 46.3 (2015-12-10) GFL (ResNet-50) — 44.5 (2015-12-10) ATSS (ResNet-50) — 43.5 (2015-12-10) FPN+ — 39.8 (2016-12-09) FPN+ — 39.8 (2016-12-09) FPN+ — 39.8 (2016-12-09) FPN+ — 39.8 (2016-12-09) FPN+ — 39.8 (2016-12-09) Mask R-CNN (ResNet-101-FPN) — 40.0 (2017-03-20) Mask R-CNN (ResNet-50-FPN) — 37.7 (2017-03-20) Mask R-CNN (ResNeXt-101-FPN) — 36.7 (2017-03-20) Mask R-CNN (ResNet-101-FPN) — 40.0 (2017-03-20) Mask R-CNN (ResNet-50-FPN) — 37.7 (2017-03-20) Mask R-CNN (ResNeXt-101-FPN) — 36.7 (2017-03-20) Mask R-CNN (ResNet-101-FPN) — 40.0 (2017-03-20) Mask R-CNN (ResNet-50-FPN) — 37.7 (2017-03-20) Mask R-CNN (ResNeXt-101-FPN) — 36.7 (2017-03-20) Mask R-CNN (ResNet-101-FPN) — 40.0 (2017-03-20) Mask R-CNN (ResNet-50-FPN) — 37.7 (2017-03-20) Mask R-CNN (ResNeXt-101-FPN) — 36.7 (2017-03-20) Mask R-CNN (ResNet-101-FPN) — 40.0 (2017-03-20) Mask R-CNN (ResNet-50-FPN) — 37.7 (2017-03-20) Mask R-CNN (ResNeXt-101-FPN) — 36.7 (2017-03-20) Mask R-CNN (ResNeXt-152 + 1 NL) — 45.0 (2017-11-21) Mask R-CNN (ResNet-101 + 1 NL) — 40.8 (2017-11-21) Mask R-CNN (ResNet-50 + 1 NL) — 39.0 (2017-11-21) Mask R-CNN (ResNeXt-152 + 1 NL) — 45.0 (2017-11-21) Mask R-CNN (ResNet-101 + 1 NL) — 40.8 (2017-11-21) Mask R-CNN (ResNet-50 + 1 NL) — 39.0 (2017-11-21) Mask R-CNN (ResNeXt-152 + 1 NL) — 45.0 (2017-11-21) Mask R-CNN (ResNet-101 + 1 NL) — 40.8 (2017-11-21) Mask R-CNN (ResNet-50 + 1 NL) — 39.0 (2017-11-21) Mask R-CNN (ResNeXt-152 + 1 NL) — 45.0 (2017-11-21) Mask R-CNN (ResNet-101 + 1 NL) — 40.8 (2017-11-21) Mask R-CNN (ResNet-50 + 1 NL) — 39.0 (2017-11-21) Mask R-CNN (ResNeXt-152 + 1 NL) — 45.0 (2017-11-21) Mask R-CNN (ResNet-101 + 1 NL) — 40.8 (2017-11-21) Mask R-CNN (ResNet-50 + 1 NL) — 39.0 (2017-11-21) Cascade R-CNN (ResNet-101-FPN+, cascade) — 42.7 (2017-12-03) Cascade R-CNN (ResNet-50-FPN+) — 40.3 (2017-12-03) Cascade R-CNN (ResNet-101-FPN+, cascade) — 42.7 (2017-12-03) Cascade R-CNN (ResNet-50-FPN+) — 40.3 (2017-12-03) Cascade R-CNN (ResNet-101-FPN+, cascade) — 42.7 (2017-12-03) Cascade R-CNN (ResNet-50-FPN+) — 40.3 (2017-12-03) Cascade R-CNN (ResNet-101-FPN+, cascade) — 42.7 (2017-12-03) Cascade R-CNN (ResNet-50-FPN+) — 40.3 (2017-12-03) Cascade R-CNN (ResNet-101-FPN+, cascade) — 42.7 (2017-12-03) Cascade R-CNN (ResNet-50-FPN+) — 40.3 (2017-12-03) Mask R-CNN (ResNet-101-FPN, GroupNorm, long) — 42.3 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm, long) — 40.8 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm) — 40.3 (2018-03-22) Mask R-CNN (ResNet-101-FPN, GroupNorm, long) — 42.3 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm, long) — 40.8 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm) — 40.3 (2018-03-22) Mask R-CNN (ResNet-101-FPN, GroupNorm, long) — 42.3 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm, long) — 40.8 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm) — 40.3 (2018-03-22) Mask R-CNN (ResNet-101-FPN, GroupNorm, long) — 42.3 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm, long) — 40.8 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm) — 40.3 (2018-03-22) Mask R-CNN (ResNet-101-FPN, GroupNorm, long) — 42.3 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm, long) — 40.8 (2018-03-22) Mask R-CNN (ResNet-50-FPN, GroupNorm) — 40.3 (2018-03-22) CornerNet511 (Hourglass-104) — 38.4 (2018-08-03) CornerNet511 (Hourglass-104) — 38.4 (2018-08-03) CornerNet511 (Hourglass-104) — 38.4 (2018-08-03) CornerNet511 (Hourglass-104) — 38.4 (2018-08-03) CornerNet511 (Hourglass-104) — 38.4 (2018-08-03) M2Det (ResNet-1o1, 320x320) — 34.1 (2018-11-12) M2Det (VGG-16, 320x320) — 33.2 (2018-11-12) M2Det (ResNet-1o1, 320x320) — 34.1 (2018-11-12) M2Det (VGG-16, 320x320) — 33.2 (2018-11-12) M2Det (ResNet-1o1, 320x320) — 34.1 (2018-11-12) M2Det (VGG-16, 320x320) — 33.2 (2018-11-12) M2Det (ResNet-1o1, 320x320) — 34.1 (2018-11-12) M2Det (VGG-16, 320x320) — 33.2 (2018-11-12) M2Det (ResNet-1o1, 320x320) — 34.1 (2018-11-12) M2Det (VGG-16, 320x320) — 33.2 (2018-11-12) GHM-C + GHM-R (RetinaNet-FPN-ResNet-50, M=30) — 35.8 (2018-11-13) GHM-C + GHM-R (RetinaNet-FPN-ResNet-50, M=30) — 35.8 (2018-11-13) GHM-C + GHM-R (RetinaNet-FPN-ResNet-50, M=30) — 35.8 (2018-11-13) GHM-C + GHM-R (RetinaNet-FPN-ResNet-50, M=30) — 35.8 (2018-11-13) GHM-C + GHM-R (RetinaNet-FPN-ResNet-50, M=30) — 35.8 (2018-11-13) Mask R-CNN (ResNeXt-152-FPN, cascade) — 48.6 (2018-11-21) Mask R-CNN (ResNet-101-FPN, GN, Cascade) — 47.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN) — 46.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN, cascade) — 48.6 (2018-11-21) Mask R-CNN (ResNet-101-FPN, GN, Cascade) — 47.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN) — 46.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN, cascade) — 48.6 (2018-11-21) Mask R-CNN (ResNet-101-FPN, GN, Cascade) — 47.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN) — 46.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN, cascade) — 48.6 (2018-11-21) Mask R-CNN (ResNet-101-FPN, GN, Cascade) — 47.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN) — 46.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN, cascade) — 48.6 (2018-11-21) Mask R-CNN (ResNet-101-FPN, GN, Cascade) — 47.4 (2018-11-21) Mask R-CNN (ResNeXt-152-FPN) — 46.4 (2018-11-21) Mask R-CNN (ResNet-101, DCNv2) — 43.1 (2018-11-27) Faster R-CNN (ResNet-101, DCNv2) — 41.7 (2018-11-27) Mask R-CNN (ResNet-101, DCNv2) — 43.1 (2018-11-27) Faster R-CNN (ResNet-101, DCNv2) — 41.7 (2018-11-27) Mask R-CNN (ResNet-101, DCNv2) — 43.1 (2018-11-27) Faster R-CNN (ResNet-101, DCNv2) — 41.7 (2018-11-27) Mask R-CNN (ResNet-101, DCNv2) — 43.1 (2018-11-27) Faster R-CNN (ResNet-101, DCNv2) — 41.7 (2018-11-27) Mask R-CNN (ResNet-101, DCNv2) — 43.1 (2018-11-27) Faster R-CNN (ResNet-101, DCNv2) — 41.7 (2018-11-27) Grid R-CNN (ResNet-101-FPN) — 41.3 (2018-11-29) Grid R-CNN (ResNet-50-FPN) — 39.6 (2018-11-29) Grid R-CNN (ResNet-101-FPN) — 41.3 (2018-11-29) Grid R-CNN (ResNet-50-FPN) — 39.6 (2018-11-29) Grid R-CNN (ResNet-101-FPN) — 41.3 (2018-11-29) Grid R-CNN (ResNet-50-FPN) — 39.6 (2018-11-29) Grid R-CNN (ResNet-101-FPN) — 41.3 (2018-11-29) Grid R-CNN (ResNet-50-FPN) — 39.6 (2018-11-29) Grid R-CNN (ResNet-101-FPN) — 41.3 (2018-11-29) Grid R-CNN (ResNet-50-FPN) — 39.6 (2018-11-29) TridentNet (ResNet-101) — 42.0 (2019-01-07) TridentNet (ResNet-101) — 42.0 (2019-01-07) TridentNet (ResNet-101) — 42.0 (2019-01-07) TridentNet (ResNet-101) — 42.0 (2019-01-07) TridentNet (ResNet-101) — 42.0 (2019-01-07) RetinaMask (ResNet-101-FPN) — 41.1 (2019-01-10) RetinaMask (ResNet-101-FPN) — 41.1 (2019-01-10) RetinaMask (ResNet-101-FPN) — 41.1 (2019-01-10) RetinaMask (ResNet-101-FPN) — 41.1 (2019-01-10) RetinaMask (ResNet-101-FPN) — 41.1 (2019-01-10) HTC (cascade) — 43.2 (2019-01-22) HTC (cascade) — 43.2 (2019-01-22) HTC (cascade) — 43.2 (2019-01-22) HTC (cascade) — 43.2 (2019-01-22) HTC (cascade) — 43.2 (2019-01-22) ExtremeNet (Hourglass-104, multi-scale) — 43.3 (2019-01-23) ExtremeNet (Hourglass-104, single-scale) — 40.3 (2019-01-23) ExtremeNet (Hourglass-104, multi-scale) — 43.3 (2019-01-23) ExtremeNet (Hourglass-104, single-scale) — 40.3 (2019-01-23) ExtremeNet (Hourglass-104, multi-scale) — 43.3 (2019-01-23) ExtremeNet (Hourglass-104, single-scale) — 40.3 (2019-01-23) ExtremeNet (Hourglass-104, multi-scale) — 43.3 (2019-01-23) ExtremeNet (Hourglass-104, single-scale) — 40.3 (2019-01-23) ExtremeNet (Hourglass-104, multi-scale) — 43.3 (2019-01-23) ExtremeNet (Hourglass-104, single-scale) — 40.3 (2019-01-23) FSAF (ResNeXt-101, anchor-based branches) — 41.6 (2019-03-02) FSAF (ResNet-101, anchor-based branches) — 39.3 (2019-03-02) FSAF (ResNet-101) — 37.9 (2019-03-02) FSAF (ResNet-50) — 35.9 (2019-03-02) FSAF (ResNeXt-101, anchor-based branches) — 41.6 (2019-03-02) FSAF (ResNet-101, anchor-based branches) — 39.3 (2019-03-02) FSAF (ResNet-101) — 37.9 (2019-03-02) FSAF (ResNet-50) — 35.9 (2019-03-02) FSAF (ResNeXt-101, anchor-based branches) — 41.6 (2019-03-02) FSAF (ResNet-101, anchor-based branches) — 39.3 (2019-03-02) FSAF (ResNet-101) — 37.9 (2019-03-02) FSAF (ResNet-50) — 35.9 (2019-03-02) FSAF (ResNeXt-101, anchor-based branches) — 41.6 (2019-03-02) FSAF (ResNet-101, anchor-based branches) — 39.3 (2019-03-02) FSAF (ResNet-101) — 37.9 (2019-03-02) FSAF (ResNet-50) — 35.9 (2019-03-02) FSAF (ResNeXt-101, anchor-based branches) — 41.6 (2019-03-02) FSAF (ResNet-101, anchor-based branches) — 39.3 (2019-03-02) FSAF (ResNet-101) — 37.9 (2019-03-02) FSAF (ResNet-50) — 35.9 (2019-03-02) Mask R-CNN-FPN (ResNeXt-101, GN+WS) — 43.12 (2019-03-25) Mask R-CNN-FPN (ResNeXt-101, GN+WS) — 43.12 (2019-03-25) Mask R-CNN-FPN (ResNeXt-101, GN+WS) — 43.12 (2019-03-25) Mask R-CNN-FPN (ResNeXt-101, GN+WS) — 43.12 (2019-03-25) Mask R-CNN-FPN (ResNeXt-101, GN+WS) — 43.12 (2019-03-25) Res2Net101+HTC — 47.5 (2019-04-02) FCOS (ResNet-50-FPN + improvements) — 38.6 (2019-04-02) Faster R-CNN (Res2Net-50) — 33.7 (2019-04-02) Res2Net101+HTC — 47.5 (2019-04-02) FCOS (ResNet-50-FPN + improvements) — 38.6 (2019-04-02) Faster R-CNN (Res2Net-50) — 33.7 (2019-04-02) Res2Net101+HTC — 47.5 (2019-04-02) FCOS (ResNet-50-FPN + improvements) — 38.6 (2019-04-02) Faster R-CNN (Res2Net-50) — 33.7 (2019-04-02) Res2Net101+HTC — 47.5 (2019-04-02) FCOS (ResNet-50-FPN + improvements) — 38.6 (2019-04-02) Faster R-CNN (Res2Net-50) — 33.7 (2019-04-02) Res2Net101+HTC — 47.5 (2019-04-02) FCOS (ResNet-50-FPN + improvements) — 38.6 (2019-04-02) Faster R-CNN (Res2Net-50) — 33.7 (2019-04-02) Libra R-CNN (ResNet-50 FPN) — 38.5 (2019-04-04) Libra R-CNN (ResNet-50 FPN) — 38.5 (2019-04-04) Libra R-CNN (ResNet-50 FPN) — 38.5 (2019-04-04) Libra R-CNN (ResNet-50 FPN) — 38.5 (2019-04-04) Libra R-CNN (ResNet-50 FPN) — 38.5 (2019-04-04) Mask R-CNN (ResNet-50, ACNet) — 39.5 (2019-04-07) Mask R-CNN (ResNet-50, ACNet) — 39.5 (2019-04-07) Mask R-CNN (ResNet-50, ACNet) — 39.5 (2019-04-07) Mask R-CNN (ResNet-50, ACNet) — 39.5 (2019-04-07) Mask R-CNN (ResNet-50, ACNet) — 39.5 (2019-04-07) FoveaBox (ResNet-101-FPN, 800x800) — 38.9 (2019-04-08) FoveaBox+Retina (ResNet-50) — 38.1 (2019-04-08) FoveaBox (ResNet-101-FPN, 600x600) — 38.0 (2019-04-08) FoveaBox (ResNet-50-FPN, 600x600) — 36.0 (2019-04-08) FoveaBox (ResNet-101-FPN, 800x800) — 38.9 (2019-04-08) FoveaBox+Retina (ResNet-50) — 38.1 (2019-04-08) FoveaBox (ResNet-101-FPN, 600x600) — 38.0 (2019-04-08) FoveaBox (ResNet-50-FPN, 600x600) — 36.0 (2019-04-08) FoveaBox (ResNet-101-FPN, 800x800) — 38.9 (2019-04-08) FoveaBox+Retina (ResNet-50) — 38.1 (2019-04-08) FoveaBox (ResNet-101-FPN, 600x600) — 38.0 (2019-04-08) FoveaBox (ResNet-50-FPN, 600x600) — 36.0 (2019-04-08) FoveaBox (ResNet-101-FPN, 800x800) — 38.9 (2019-04-08) FoveaBox+Retina (ResNet-50) — 38.1 (2019-04-08) FoveaBox (ResNet-101-FPN, 600x600) — 38.0 (2019-04-08) FoveaBox (ResNet-50-FPN, 600x600) — 36.0 (2019-04-08) FoveaBox (ResNet-101-FPN, 800x800) — 38.9 (2019-04-08) FoveaBox+Retina (ResNet-50) — 38.1 (2019-04-08) FoveaBox (ResNet-101-FPN, 600x600) — 38.0 (2019-04-08) FoveaBox (ResNet-50-FPN, 600x600) — 36.0 (2019-04-08) CenterNet511 (Hourglass-52) — 41.3 (2019-04-17) CenterNet511 (Hourglass-52) — 41.3 (2019-04-17) CenterNet511 (Hourglass-52) — 41.3 (2019-04-17) CenterNet511 (Hourglass-52) — 41.3 (2019-04-17) CenterNet511 (Hourglass-52) — 41.3 (2019-04-17) CornerNet-Saccade (Hourglass-54) — 42.6 (2019-04-18) CornerNet-Saccade (Hourglass-104) — 41.4 (2019-04-18) CornerNet-Saccade (Hourglass-54) — 42.6 (2019-04-18) CornerNet-Saccade (Hourglass-104) — 41.4 (2019-04-18) CornerNet-Saccade (Hourglass-54) — 42.6 (2019-04-18) CornerNet-Saccade (Hourglass-104) — 41.4 (2019-04-18) CornerNet-Saccade (Hourglass-54) — 42.6 (2019-04-18) CornerNet-Saccade (Hourglass-104) — 41.4 (2019-04-18) CornerNet-Saccade (Hourglass-54) — 42.6 (2019-04-18) CornerNet-Saccade (Hourglass-104) — 41.4 (2019-04-18) GCNet (ResNeXt-101 + DCN + cascade + GC r16) — 47.9 (2019-04-25) RPDet (ResNeXt-101-DCN, multi-scale) — 46.8 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale) — 46.4 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale train) — 44.8 (2019-04-25) RPDet (ResNeXt-101-DCN) — 44.5 (2019-04-25) RPDet (ResNet-50, multi-scale train) — 40.8 (2019-04-25) GCnet (ResNet-50-FPN, GRoIE) — 40.3 (2019-04-25) RPDet (ResNet-101) — 40.3 (2019-04-25) RPDet (ResNet-50) — 38.6 (2019-04-25) GCNet (ResNeXt-101 + DCN + cascade + GC r16) — 47.9 (2019-04-25) RPDet (ResNeXt-101-DCN, multi-scale) — 46.8 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale) — 46.4 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale train) — 44.8 (2019-04-25) RPDet (ResNeXt-101-DCN) — 44.5 (2019-04-25) RPDet (ResNet-50, multi-scale train) — 40.8 (2019-04-25) GCnet (ResNet-50-FPN, GRoIE) — 40.3 (2019-04-25) RPDet (ResNet-101) — 40.3 (2019-04-25) RPDet (ResNet-50) — 38.6 (2019-04-25) GCNet (ResNeXt-101 + DCN + cascade + GC r16) — 47.9 (2019-04-25) RPDet (ResNeXt-101-DCN, multi-scale) — 46.8 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale) — 46.4 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale train) — 44.8 (2019-04-25) RPDet (ResNeXt-101-DCN) — 44.5 (2019-04-25) RPDet (ResNet-50, multi-scale train) — 40.8 (2019-04-25) GCnet (ResNet-50-FPN, GRoIE) — 40.3 (2019-04-25) RPDet (ResNet-101) — 40.3 (2019-04-25) RPDet (ResNet-50) — 38.6 (2019-04-25) GCNet (ResNeXt-101 + DCN + cascade + GC r16) — 47.9 (2019-04-25) RPDet (ResNeXt-101-DCN, multi-scale) — 46.8 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale) — 46.4 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale train) — 44.8 (2019-04-25) RPDet (ResNeXt-101-DCN) — 44.5 (2019-04-25) RPDet (ResNet-50, multi-scale train) — 40.8 (2019-04-25) GCnet (ResNet-50-FPN, GRoIE) — 40.3 (2019-04-25) RPDet (ResNet-101) — 40.3 (2019-04-25) RPDet (ResNet-50) — 38.6 (2019-04-25) GCNet (ResNeXt-101 + DCN + cascade + GC r16) — 47.9 (2019-04-25) RPDet (ResNeXt-101-DCN, multi-scale) — 46.8 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale) — 46.4 (2019-04-25) RPDet (ResNet-101-DCN, multi-scale train) — 44.8 (2019-04-25) RPDet (ResNeXt-101-DCN) — 44.5 (2019-04-25) RPDet (ResNet-50, multi-scale train) — 40.8 (2019-04-25) GCnet (ResNet-50-FPN, GRoIE) — 40.3 (2019-04-25) RPDet (ResNet-101) — 40.3 (2019-04-25) RPDet (ResNet-50) — 38.6 (2019-04-25) Mask R-CNN-FPN (AOGNet-40M) — 44.9 (2019-08-04) Mask R-CNN-FPN (AOGNet-40M) — 44.9 (2019-08-04) Mask R-CNN-FPN (AOGNet-40M) — 44.9 (2019-08-04) Mask R-CNN-FPN (AOGNet-40M) — 44.9 (2019-08-04) Mask R-CNN-FPN (AOGNet-40M) — 44.9 (2019-08-04) Faster R-CNN (LIP-ResNet-101) — 41.7 (2019-08-12) Faster R-CNN (LIP-ResNet-101) — 41.7 (2019-08-12) Faster R-CNN (LIP-ResNet-101) — 41.7 (2019-08-12) Faster R-CNN (LIP-ResNet-101) — 41.7 (2019-08-12) Faster R-CNN (LIP-ResNet-101) — 41.7 (2019-08-12) HTC (HRNetV2p-W48) — 47.0 (2019-08-20) Mask R-CNN (HRNetV2p-W48, cascade) — 46.0 (2019-08-20) HTC (HRNetV2p-W32) — 45.3 (2019-08-20) Cascade R-CNN (HRNetV2p-W48) — 44.6 (2019-08-20) Cascade R-CNN (HRNetV2p-W32) — 43.7 (2019-08-20) HTC (HRNetV2p-W18) — 43.1 (2019-08-20) Mask R-CNN (HRNetV2p-W32) — 42.3 (2019-08-20) Faster R-CNN (HRNetV2p-W48) — 41.8 (2019-08-20) Cascade R-CNN (HRNetV2p-W18) — 41.3 (2019-08-20) Faster R-CNN (HRNetV2p-W32) — 40.9 (2019-08-20) Mask R-CNN (HRNetV2p-W18) — 39.2 (2019-08-20) Faster R-CNN (HRNetV2p-W18) — 38.0 (2019-08-20) HTC (HRNetV2p-W48) — 47.0 (2019-08-20) Mask R-CNN (HRNetV2p-W48, cascade) — 46.0 (2019-08-20) HTC (HRNetV2p-W32) — 45.3 (2019-08-20) Cascade R-CNN (HRNetV2p-W48) — 44.6 (2019-08-20) Cascade R-CNN (HRNetV2p-W32) — 43.7 (2019-08-20) HTC (HRNetV2p-W18) — 43.1 (2019-08-20) Mask R-CNN (HRNetV2p-W32) — 42.3 (2019-08-20) Faster R-CNN (HRNetV2p-W48) — 41.8 (2019-08-20) Cascade R-CNN (HRNetV2p-W18) — 41.3 (2019-08-20) Faster R-CNN (HRNetV2p-W32) — 40.9 (2019-08-20) Mask R-CNN (HRNetV2p-W18) — 39.2 (2019-08-20) Faster R-CNN (HRNetV2p-W18) — 38.0 (2019-08-20) HTC (HRNetV2p-W48) — 47.0 (2019-08-20) Mask R-CNN (HRNetV2p-W48, cascade) — 46.0 (2019-08-20) HTC (HRNetV2p-W32) — 45.3 (2019-08-20) Cascade R-CNN (HRNetV2p-W48) — 44.6 (2019-08-20) Cascade R-CNN (HRNetV2p-W32) — 43.7 (2019-08-20) HTC (HRNetV2p-W18) — 43.1 (2019-08-20) Mask R-CNN (HRNetV2p-W32) — 42.3 (2019-08-20) Faster R-CNN (HRNetV2p-W48) — 41.8 (2019-08-20) Cascade R-CNN (HRNetV2p-W18) — 41.3 (2019-08-20) Faster R-CNN (HRNetV2p-W32) — 40.9 (2019-08-20) Mask R-CNN (HRNetV2p-W18) — 39.2 (2019-08-20) Faster R-CNN (HRNetV2p-W18) — 38.0 (2019-08-20) HTC (HRNetV2p-W48) — 47.0 (2019-08-20) Mask R-CNN (HRNetV2p-W48, cascade) — 46.0 (2019-08-20) HTC (HRNetV2p-W32) — 45.3 (2019-08-20) Cascade R-CNN (HRNetV2p-W48) — 44.6 (2019-08-20) Cascade R-CNN (HRNetV2p-W32) — 43.7 (2019-08-20) HTC (HRNetV2p-W18) — 43.1 (2019-08-20) Mask R-CNN (HRNetV2p-W32) — 42.3 (2019-08-20) Faster R-CNN (HRNetV2p-W48) — 41.8 (2019-08-20) Cascade R-CNN (HRNetV2p-W18) — 41.3 (2019-08-20) Faster R-CNN (HRNetV2p-W32) — 40.9 (2019-08-20) Mask R-CNN (HRNetV2p-W18) — 39.2 (2019-08-20) Faster R-CNN (HRNetV2p-W18) — 38.0 (2019-08-20) HTC (HRNetV2p-W48) — 47.0 (2019-08-20) Mask R-CNN (HRNetV2p-W48, cascade) — 46.0 (2019-08-20) HTC (HRNetV2p-W32) — 45.3 (2019-08-20) Cascade R-CNN (HRNetV2p-W48) — 44.6 (2019-08-20) Cascade R-CNN (HRNetV2p-W32) — 43.7 (2019-08-20) HTC (HRNetV2p-W18) — 43.1 (2019-08-20) Mask R-CNN (HRNetV2p-W32) — 42.3 (2019-08-20) Faster R-CNN (HRNetV2p-W48) — 41.8 (2019-08-20) Cascade R-CNN (HRNetV2p-W18) — 41.3 (2019-08-20) Faster R-CNN (HRNetV2p-W32) — 40.9 (2019-08-20) Mask R-CNN (HRNetV2p-W18) — 39.2 (2019-08-20) Faster R-CNN (HRNetV2p-W18) — 38.0 (2019-08-20) Online Fg Bal. Sampling+Hard Negative Mining (ResNet-50) — 35.6 (2019-09-21) Online Fg Bal. Sampling+Hard Negative Mining (ResNet-50) — 35.6 (2019-09-21) Online Fg Bal. Sampling+Hard Negative Mining (ResNet-50) — 35.6 (2019-09-21) Online Fg Bal. Sampling+Hard Negative Mining (ResNet-50) — 35.6 (2019-09-21) Online Fg Bal. Sampling+Hard Negative Mining (ResNet-50) — 35.6 (2019-09-21) CenterMask+VoVNet99 (multi-scale) — 48.6 (2019-11-15) CenterMask+VoVNetV2-99 (single-scale) — 45.6 (2019-11-15) Mask R-CNN (VoVNetV2-99, single-scale) — 44.9 (2019-11-15) CenterMask+VoVNetV2-57 (single-scale) — 44.6 (2019-11-15) CenterMask+X101-32x8d (single-scale) — 44.4 (2019-11-15) CenterMask+VoVNet99 (multi-scale) — 48.6 (2019-11-15) CenterMask+VoVNetV2-99 (single-scale) — 45.6 (2019-11-15) Mask R-CNN (VoVNetV2-99, single-scale) — 44.9 (2019-11-15) CenterMask+VoVNetV2-57 (single-scale) — 44.6 (2019-11-15) CenterMask+X101-32x8d (single-scale) — 44.4 (2019-11-15) CenterMask+VoVNet99 (multi-scale) — 48.6 (2019-11-15) CenterMask+VoVNetV2-99 (single-scale) — 45.6 (2019-11-15) Mask R-CNN (VoVNetV2-99, single-scale) — 44.9 (2019-11-15) CenterMask+VoVNetV2-57 (single-scale) — 44.6 (2019-11-15) CenterMask+X101-32x8d (single-scale) — 44.4 (2019-11-15) CenterMask+VoVNet99 (multi-scale) — 48.6 (2019-11-15) CenterMask+VoVNetV2-99 (single-scale) — 45.6 (2019-11-15) Mask R-CNN (VoVNetV2-99, single-scale) — 44.9 (2019-11-15) CenterMask+VoVNetV2-57 (single-scale) — 44.6 (2019-11-15) CenterMask+X101-32x8d (single-scale) — 44.4 (2019-11-15) CenterMask+VoVNet99 (multi-scale) — 48.6 (2019-11-15) CenterMask+VoVNetV2-99 (single-scale) — 45.6 (2019-11-15) Mask R-CNN (VoVNetV2-99, single-scale) — 44.9 (2019-11-15) CenterMask+VoVNetV2-57 (single-scale) — 44.6 (2019-11-15) CenterMask+X101-32x8d (single-scale) — 44.4 (2019-11-15) EfficientDet-D7 (1536) — 52.1 (2019-11-20) EfficientDet-D7 (1536) — 52.1 (2019-11-20) EfficientDet-D7 (1536) — 52.1 (2019-11-20) EfficientDet-D7 (1536) — 52.1 (2019-11-20) EfficientDet-D7 (1536) — 52.1 (2019-11-20) RetinaNet (SpineNet-190, 1536x1536) — 52.2 (2019-12-10) RetinaNet (SpineNet-190, 1536x1536) — 52.2 (2019-12-10) RetinaNet (SpineNet-190, 1536x1536) — 52.2 (2019-12-10) RetinaNet (SpineNet-190, 1536x1536) — 52.2 (2019-12-10) RetinaNet (SpineNet-190, 1536x1536) — 52.2 (2019-12-10) ResNeSt-200 (multi-scale) — 52.47 (2020-04-19) ResNeSt-200-DCN (single-scale) — 50.91 (2020-04-19) ResNeSt-200 (single-scale) — 50.54 (2020-04-19) ResNeSt-200 (multi-scale) — 52.47 (2020-04-19) ResNeSt-200-DCN (single-scale) — 50.91 (2020-04-19) ResNeSt-200 (single-scale) — 50.54 (2020-04-19) ResNeSt-200 (multi-scale) — 52.47 (2020-04-19) ResNeSt-200-DCN (single-scale) — 50.91 (2020-04-19) ResNeSt-200 (single-scale) — 50.54 (2020-04-19) ResNeSt-200 (multi-scale) — 52.47 (2020-04-19) ResNeSt-200-DCN (single-scale) — 50.91 (2020-04-19) ResNeSt-200 (single-scale) — 50.54 (2020-04-19) ResNeSt-200 (multi-scale) — 52.47 (2020-04-19) ResNeSt-200-DCN (single-scale) — 50.91 (2020-04-19) ResNeSt-200 (single-scale) — 50.54 (2020-04-19) Mask R-CNN (ResNet-50-FPN, GRoIE) — 38.4 (2020-04-28) Faster R-CNN (ResNet-50-FPN, GRoIE) — 37.5 (2020-04-28) Mask R-CNN (ResNet-50-FPN, GRoIE) — 38.4 (2020-04-28) Faster R-CNN (ResNet-50-FPN, GRoIE) — 37.5 (2020-04-28) Mask R-CNN (ResNet-50-FPN, GRoIE) — 38.4 (2020-04-28) Faster R-CNN (ResNet-50-FPN, GRoIE) — 37.5 (2020-04-28) Mask R-CNN (ResNet-50-FPN, GRoIE) — 38.4 (2020-04-28) Faster R-CNN (ResNet-50-FPN, GRoIE) — 37.5 (2020-04-28) Mask R-CNN (ResNet-50-FPN, GRoIE) — 38.4 (2020-04-28) Faster R-CNN (ResNet-50-FPN, GRoIE) — 37.5 (2020-04-28) DETR-DC5 (ResNet-101) — 44.9 (2020-05-26) Faster RCNN-R101-FPN+ — 44.0 (2020-05-26) DETR-DC5 (ResNet-101) — 44.9 (2020-05-26) Faster RCNN-R101-FPN+ — 44.0 (2020-05-26) DETR-DC5 (ResNet-101) — 44.9 (2020-05-26) Faster RCNN-R101-FPN+ — 44.0 (2020-05-26) DETR-DC5 (ResNet-101) — 44.9 (2020-05-26) Faster RCNN-R101-FPN+ — 44.0 (2020-05-26) DETR-DC5 (ResNet-101) — 44.9 (2020-05-26) Faster RCNN-R101-FPN+ — 44.0 (2020-05-26) SpineNet-190 (1280, with Self-training on OpenImages, single-scale) — 54.2 (2020-06-11) VirTex Mask R-CNN (ResNet-50-FPN) — 40.9 (2020-06-11) SpineNet-190 (1280, with Self-training on OpenImages, single-scale) — 54.2 (2020-06-11) VirTex Mask R-CNN (ResNet-50-FPN) — 40.9 (2020-06-11) SpineNet-190 (1280, with Self-training on OpenImages, single-scale) — 54.2 (2020-06-11) VirTex Mask R-CNN (ResNet-50-FPN) — 40.9 (2020-06-11) SpineNet-190 (1280, with Self-training on OpenImages, single-scale) — 54.2 (2020-06-11) VirTex Mask R-CNN (ResNet-50-FPN) — 40.9 (2020-06-11) SpineNet-190 (1280, with Self-training on OpenImages, single-scale) — 54.2 (2020-06-11) VirTex Mask R-CNN (ResNet-50-FPN) — 40.9 (2020-06-11) HoughNet (HG-104, MS) — 46.1 (2020-07-05) HoughNet (HG-104) — 43.0 (2020-07-05) HoughNet (HG-104, MS) — 46.1 (2020-07-05) HoughNet (HG-104) — 43.0 (2020-07-05) HoughNet (HG-104, MS) — 46.1 (2020-07-05) HoughNet (HG-104) — 43.0 (2020-07-05) HoughNet (HG-104, MS) — 46.1 (2020-07-05) HoughNet (HG-104) — 43.0 (2020-07-05) HoughNet (HG-104, MS) — 46.1 (2020-07-05) HoughNet (HG-104) — 43.0 (2020-07-05) PPDet (ResNet-101-FPN) — 40.5 (2020-08-03) PPDet (ResNet-101-FPN) — 40.5 (2020-08-03) PPDet (ResNet-101-FPN) — 40.5 (2020-08-03) PPDet (ResNet-101-FPN) — 40.5 (2020-08-03) PPDet (ResNet-101-FPN) — 40.5 (2020-08-03) Faster R-CNN+aLRP Loss (ResNet-50, 500 scale) — 40.7 (2020-09-28) RetinaNet+aLRP Loss (ResNet-50, 500 scale) — 40.2 (2020-09-28) FoveaBox+aLRP Loss (ResNet-50, 500 scale) — 39.7 (2020-09-28) Faster R-CNN+aLRP Loss (ResNet-50, 500 scale) — 40.7 (2020-09-28) RetinaNet+aLRP Loss (ResNet-50, 500 scale) — 40.2 (2020-09-28) FoveaBox+aLRP Loss (ResNet-50, 500 scale) — 39.7 (2020-09-28) Faster R-CNN+aLRP Loss (ResNet-50, 500 scale) — 40.7 (2020-09-28) RetinaNet+aLRP Loss (ResNet-50, 500 scale) — 40.2 (2020-09-28) FoveaBox+aLRP Loss (ResNet-50, 500 scale) — 39.7 (2020-09-28) Faster R-CNN+aLRP Loss (ResNet-50, 500 scale) — 40.7 (2020-09-28) RetinaNet+aLRP Loss (ResNet-50, 500 scale) — 40.2 (2020-09-28) FoveaBox+aLRP Loss (ResNet-50, 500 scale) — 39.7 (2020-09-28) Faster R-CNN+aLRP Loss (ResNet-50, 500 scale) — 40.7 (2020-09-28) RetinaNet+aLRP Loss (ResNet-50, 500 scale) — 40.2 (2020-09-28) FoveaBox+aLRP Loss (ResNet-50, 500 scale) — 39.7 (2020-09-28) YOLOv4-P7 CSP-P7 (single-scale, 16 fps) — 55.4 (2020-11-16) YOLOv4-P7 CSP-P7 (single-scale, 16 fps) — 55.4 (2020-11-16) YOLOv4-P7 CSP-P7 (single-scale, 16 fps) — 55.4 (2020-11-16) YOLOv4-P7 CSP-P7 (single-scale, 16 fps) — 55.4 (2020-11-16) YOLOv4-P7 CSP-P7 (single-scale, 16 fps) — 55.4 (2020-11-16) Sparse R-CNN (ResNet-101, learnable proposals, random crop aug, FPN) — 45.6 (2020-11-25) Sparse R-CNN (ResNet-50, learnable proposals, random crop aug, FPN) — 44.5 (2020-11-25) Sparse R-CNN (ResNet-101, FPN) — 43.5 (2020-11-25) Sparse R-CNN (ResNet-50, FPN) — 42.3 (2020-11-25) Sparse R-CNN (ResNet-101, learnable proposals, random crop aug, FPN) — 45.6 (2020-11-25) Sparse R-CNN (ResNet-50, learnable proposals, random crop aug, FPN) — 44.5 (2020-11-25) Sparse R-CNN (ResNet-101, FPN) — 43.5 (2020-11-25) Sparse R-CNN (ResNet-50, FPN) — 42.3 (2020-11-25) Sparse R-CNN (ResNet-101, learnable proposals, random crop aug, FPN) — 45.6 (2020-11-25) Sparse R-CNN (ResNet-50, learnable proposals, random crop aug, FPN) — 44.5 (2020-11-25) Sparse R-CNN (ResNet-101, FPN) — 43.5 (2020-11-25) Sparse R-CNN (ResNet-50, FPN) — 42.3 (2020-11-25) Sparse R-CNN (ResNet-101, learnable proposals, random crop aug, FPN) — 45.6 (2020-11-25) Sparse R-CNN (ResNet-50, learnable proposals, random crop aug, FPN) — 44.5 (2020-11-25) Sparse R-CNN (ResNet-101, FPN) — 43.5 (2020-11-25) Sparse R-CNN (ResNet-50, FPN) — 42.3 (2020-11-25) Sparse R-CNN (ResNet-101, learnable proposals, random crop aug, FPN) — 45.6 (2020-11-25) Sparse R-CNN (ResNet-50, learnable proposals, random crop aug, FPN) — 44.5 (2020-11-25) Sparse R-CNN (ResNet-101, FPN) — 43.5 (2020-11-25) Sparse R-CNN (ResNet-50, FPN) — 42.3 (2020-11-25) Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) — 57.0 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280) — 54.5 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) — 57.0 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280) — 54.5 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) — 57.0 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280) — 54.5 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) — 57.0 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280) — 54.5 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) — 57.0 (2020-12-13) Cascade Eff-B7 NAS-FPN (1280) — 54.5 (2020-12-13) GCNet (ResNeXt-101 + DCN + cascade + GC r4) — 51.8 (2020-12-24) GCNet (ResNeXt-101 + DCN + cascade + GC r4) — 51.8 (2020-12-24) GCNet (ResNeXt-101 + DCN + cascade + GC r4) — 51.8 (2020-12-24) GCNet (ResNeXt-101 + DCN + cascade + GC r4) — 51.8 (2020-12-24) GCNet (ResNeXt-101 + DCN + cascade + GC r4) — 51.8 (2020-12-24) BoTNet 200 (Mask R-CNN, single scale, 72 epochs) — 49.7 (2021-01-27) BoTNet 152 (Mask R-CNN, single scale, 72 epochs) — 49.5 (2021-01-27) BoTNet 50 (72 epochs) — 45.9 (2021-01-27) BoTNet 200 (Mask R-CNN, single scale, 72 epochs) — 49.7 (2021-01-27) BoTNet 152 (Mask R-CNN, single scale, 72 epochs) — 49.5 (2021-01-27) BoTNet 50 (72 epochs) — 45.9 (2021-01-27) BoTNet 200 (Mask R-CNN, single scale, 72 epochs) — 49.7 (2021-01-27) BoTNet 152 (Mask R-CNN, single scale, 72 epochs) — 49.5 (2021-01-27) BoTNet 50 (72 epochs) — 45.9 (2021-01-27) BoTNet 200 (Mask R-CNN, single scale, 72 epochs) — 49.7 (2021-01-27) BoTNet 152 (Mask R-CNN, single scale, 72 epochs) — 49.5 (2021-01-27) BoTNet 50 (72 epochs) — 45.9 (2021-01-27) BoTNet 200 (Mask R-CNN, single scale, 72 epochs) — 49.7 (2021-01-27) BoTNet 152 (Mask R-CNN, single scale, 72 epochs) — 49.5 (2021-01-27) BoTNet 50 (72 epochs) — 45.9 (2021-01-27) PVT-Large (RetinaNet 3x,MS) — 43.4 (2021-02-24) PVT-Large (RetinaNet 1x) — 42.6 (2021-02-24) PVT-Large (RetinaNet 3x,MS) — 43.4 (2021-02-24) PVT-Large (RetinaNet 1x) — 42.6 (2021-02-24) PVT-Large (RetinaNet 3x,MS) — 43.4 (2021-02-24) PVT-Large (RetinaNet 1x) — 42.6 (2021-02-24) PVT-Large (RetinaNet 3x,MS) — 43.4 (2021-02-24) PVT-Large (RetinaNet 1x) — 42.6 (2021-02-24) PVT-Large (RetinaNet 3x,MS) — 43.4 (2021-02-24) PVT-Large (RetinaNet 1x) — 42.6 (2021-02-24) Swin-L (HTC++, multi scale) — 58.0 (2021-03-25) Swin-L (HTC++, single scale) — 57.1 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, multi-scale) — 53.5 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, single-scale) — 50.9 (2021-03-25) UniverseNet-20.08 (Res2Net-50, DCN, single-scale) — 48.5 (2021-03-25) Swin-L (HTC++, multi scale) — 58.0 (2021-03-25) Swin-L (HTC++, single scale) — 57.1 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, multi-scale) — 53.5 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, single-scale) — 50.9 (2021-03-25) UniverseNet-20.08 (Res2Net-50, DCN, single-scale) — 48.5 (2021-03-25) Swin-L (HTC++, multi scale) — 58.0 (2021-03-25) Swin-L (HTC++, single scale) — 57.1 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, multi-scale) — 53.5 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, single-scale) — 50.9 (2021-03-25) UniverseNet-20.08 (Res2Net-50, DCN, single-scale) — 48.5 (2021-03-25) Swin-L (HTC++, multi scale) — 58.0 (2021-03-25) Swin-L (HTC++, single scale) — 57.1 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, multi-scale) — 53.5 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, single-scale) — 50.9 (2021-03-25) UniverseNet-20.08 (Res2Net-50, DCN, single-scale) — 48.5 (2021-03-25) Swin-L (HTC++, multi scale) — 58.0 (2021-03-25) Swin-L (HTC++, single scale) — 57.1 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, multi-scale) — 53.5 (2021-03-25) UniverseNet-20.08d (Res2Net-101, DCN, single-scale) — 50.9 (2021-03-25) UniverseNet-20.08 (Res2Net-50, DCN, single-scale) — 48.5 (2021-03-25) RetinaNet (ViL-Base, multi-scale, 3x) — 44.7 (2021-03-29) RetinaNet (ViL-Base) — 44.3 (2021-03-29) RetinaNet (ViL-Base, multi-scale, 3x) — 44.7 (2021-03-29) RetinaNet (ViL-Base) — 44.3 (2021-03-29) RetinaNet (ViL-Base, multi-scale, 3x) — 44.7 (2021-03-29) RetinaNet (ViL-Base) — 44.3 (2021-03-29) RetinaNet (ViL-Base, multi-scale, 3x) — 44.7 (2021-03-29) RetinaNet (ViL-Base) — 44.3 (2021-03-29) RetinaNet (ViL-Base, multi-scale, 3x) — 44.7 (2021-03-29) RetinaNet (ViL-Base) — 44.3 (2021-03-29) R3-CNN (ResNet-50-FPN, DCN) — 44.8 (2021-04-03) R3-CNN (ResNet-50-FPN, GC-Net) — 44.3 (2021-04-03) R3-CNN (ResNet-50-FPN) — 42.0 (2021-04-03) R3-CNN (ResNet-50-FPN, DCN) — 44.8 (2021-04-03) R3-CNN (ResNet-50-FPN, GC-Net) — 44.3 (2021-04-03) R3-CNN (ResNet-50-FPN) — 42.0 (2021-04-03) R3-CNN (ResNet-50-FPN, DCN) — 44.8 (2021-04-03) R3-CNN (ResNet-50-FPN, GC-Net) — 44.3 (2021-04-03) R3-CNN (ResNet-50-FPN) — 42.0 (2021-04-03) R3-CNN (ResNet-50-FPN, DCN) — 44.8 (2021-04-03) R3-CNN (ResNet-50-FPN, GC-Net) — 44.3 (2021-04-03) R3-CNN (ResNet-50-FPN) — 42.0 (2021-04-03) R3-CNN (ResNet-50-FPN, DCN) — 44.8 (2021-04-03) R3-CNN (ResNet-50-FPN, GC-Net) — 44.3 (2021-04-03) R3-CNN (ResNet-50-FPN) — 42.0 (2021-04-03) QueryInst (single scale) — 56.1 (2021-05-05) QueryInst (single scale) — 56.1 (2021-05-05) QueryInst (single scale) — 56.1 (2021-05-05) QueryInst (single scale) — 56.1 (2021-05-05) QueryInst (single scale) — 56.1 (2021-05-05) Faster R-CNN (FPN, X-volution) — 42.8 (2021-06-04) Faster R-CNN (FPN, X-volution) — 42.8 (2021-06-04) Faster R-CNN (FPN, X-volution) — 42.8 (2021-06-04) Faster R-CNN (FPN, X-volution) — 42.8 (2021-06-04) Faster R-CNN (FPN, X-volution) — 42.8 (2021-06-04) DyHead (Swin-L, multi scale, self-training) — 60.3 (2021-06-15) DyHead (Swin-L, multi scale) — 58.4 (2021-06-15) DyHead (ResNet-101) — 46.5 (2021-06-15) DyHead (Swin-L, multi scale, self-training) — 60.3 (2021-06-15) DyHead (Swin-L, multi scale) — 58.4 (2021-06-15) DyHead (ResNet-101) — 46.5 (2021-06-15) DyHead (Swin-L, multi scale, self-training) — 60.3 (2021-06-15) DyHead (Swin-L, multi scale) — 58.4 (2021-06-15) DyHead (ResNet-101) — 46.5 (2021-06-15) DyHead (Swin-L, multi scale, self-training) — 60.3 (2021-06-15) DyHead (Swin-L, multi scale) — 58.4 (2021-06-15) DyHead (ResNet-101) — 46.5 (2021-06-15) DyHead (Swin-L, multi scale, self-training) — 60.3 (2021-06-15) DyHead (Swin-L, multi scale) — 58.4 (2021-06-15) DyHead (ResNet-101) — 46.5 (2021-06-15) Soft Teacher + Swin-L (HTC++, multi-scale) — 60.7 (2021-06-16) Soft Teacher+Swin-L(HTC++, single scale) — 60.1 (2021-06-16) Soft Teacher + Swin-L (HTC++, multi-scale) — 60.7 (2021-06-16) Soft Teacher+Swin-L(HTC++, single scale) — 60.1 (2021-06-16) Soft Teacher + Swin-L (HTC++, multi-scale) — 60.7 (2021-06-16) Soft Teacher+Swin-L(HTC++, single scale) — 60.1 (2021-06-16) Soft Teacher + Swin-L (HTC++, multi-scale) — 60.7 (2021-06-16) Soft Teacher+Swin-L(HTC++, single scale) — 60.1 (2021-06-16) Soft Teacher + Swin-L (HTC++, multi-scale) — 60.7 (2021-06-16) Soft Teacher+Swin-L(HTC++, single scale) — 60.1 (2021-06-16) XCiT-M24/8 — 48.5 (2021-06-17) XCiT-S24/8 — 48.1 (2021-06-17) XCiT-M24/8 — 48.5 (2021-06-17) XCiT-S24/8 — 48.1 (2021-06-17) XCiT-M24/8 — 48.5 (2021-06-17) XCiT-S24/8 — 48.1 (2021-06-17) XCiT-M24/8 — 48.5 (2021-06-17) XCiT-S24/8 — 48.1 (2021-06-17) XCiT-M24/8 — 48.5 (2021-06-17) XCiT-S24/8 — 48.1 (2021-06-17) Sparse R-CNN (PVTv2-B2) — 50.1 (2021-06-25) Sparse R-CNN (PVTv2-B2) — 50.1 (2021-06-25) Sparse R-CNN (PVTv2-B2) — 50.1 (2021-06-25) Sparse R-CNN (PVTv2-B2) — 50.1 (2021-06-25) Sparse R-CNN (PVTv2-B2) — 50.1 (2021-06-25) Cascade RCNN-RS (SpineNet-143L, single scale) — 53.6 (2021-06-30) Cascade RCNN-RS (ResNet-200, single scale) — 53.1 (2021-06-30) Cascade RCNN-RS (SpineNet-143L, single scale) — 53.6 (2021-06-30) Cascade RCNN-RS (ResNet-200, single scale) — 53.1 (2021-06-30) Cascade RCNN-RS (SpineNet-143L, single scale) — 53.6 (2021-06-30) Cascade RCNN-RS (ResNet-200, single scale) — 53.1 (2021-06-30) Cascade RCNN-RS (SpineNet-143L, single scale) — 53.6 (2021-06-30) Cascade RCNN-RS (ResNet-200, single scale) — 53.1 (2021-06-30) Cascade RCNN-RS (SpineNet-143L, single scale) — 53.6 (2021-06-30) Cascade RCNN-RS (ResNet-200, single scale) — 53.1 (2021-06-30) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.6 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.1 (2021-07-01) Focal-L (DyHead, multi-scale) — 58.7 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.6 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.1 (2021-07-01) Focal-L (DyHead, multi-scale) — 58.7 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.6 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.1 (2021-07-01) Focal-L (DyHead, multi-scale) — 58.7 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.6 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.1 (2021-07-01) Focal-L (DyHead, multi-scale) — 58.7 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.6 (2021-07-01) CBNetV2 (Dual-Swin-L HTC, multi-scale) — 59.1 (2021-07-01) Focal-L (DyHead, multi-scale) — 58.7 (2021-07-01) DETR-ResNet50 with iRPE-K (300 epochs) — 42.3 (2021-07-29) DETR-ResNet50 with iRPE-K (150 epochs) — 40.8 (2021-07-29) DETR-ResNet50 with iRPE-K (300 epochs) — 42.3 (2021-07-29) DETR-ResNet50 with iRPE-K (150 epochs) — 40.8 (2021-07-29) DETR-ResNet50 with iRPE-K (300 epochs) — 42.3 (2021-07-29) DETR-ResNet50 with iRPE-K (150 epochs) — 40.8 (2021-07-29) DETR-ResNet50 with iRPE-K (300 epochs) — 42.3 (2021-07-29) DETR-ResNet50 with iRPE-K (150 epochs) — 40.8 (2021-07-29) DETR-ResNet50 with iRPE-K (300 epochs) — 42.3 (2021-07-29) DETR-ResNet50 with iRPE-K (150 epochs) — 40.8 (2021-07-29) Conditional DETR-DC5-R101 — 45.9 (2021-08-13) Conditional DETR-DC5-R50 — 45.1 (2021-08-13) Conditional DETR-R101 — 44.5 (2021-08-13) Conditional DETR-R50 — 43.0 (2021-08-13) Conditional DETR-DC5-R101 — 45.9 (2021-08-13) Conditional DETR-DC5-R50 — 45.1 (2021-08-13) Conditional DETR-R101 — 44.5 (2021-08-13) Conditional DETR-R50 — 43.0 (2021-08-13) Conditional DETR-DC5-R101 — 45.9 (2021-08-13) Conditional DETR-DC5-R50 — 45.1 (2021-08-13) Conditional DETR-R101 — 44.5 (2021-08-13) Conditional DETR-R50 — 43.0 (2021-08-13) Conditional DETR-DC5-R101 — 45.9 (2021-08-13) Conditional DETR-DC5-R50 — 45.1 (2021-08-13) Conditional DETR-R101 — 44.5 (2021-08-13) Conditional DETR-R50 — 43.0 (2021-08-13) Conditional DETR-DC5-R101 — 45.9 (2021-08-13) Conditional DETR-DC5-R50 — 45.1 (2021-08-13) Conditional DETR-R101 — 44.5 (2021-08-13) Conditional DETR-R50 — 43.0 (2021-08-13) Anchor DETR-DC5-R101 — 45.1 (2021-09-15) Anchor DETR-DC5-R50 — 44.2 (2021-09-15) Anchor DETR-DC5-R101 — 45.1 (2021-09-15) Anchor DETR-DC5-R50 — 44.2 (2021-09-15) Anchor DETR-DC5-R101 — 45.1 (2021-09-15) Anchor DETR-DC5-R50 — 44.2 (2021-09-15) Anchor DETR-DC5-R101 — 45.1 (2021-09-15) Anchor DETR-DC5-R50 — 44.2 (2021-09-15) Anchor DETR-DC5-R101 — 45.1 (2021-09-15) Anchor DETR-DC5-R50 — 44.2 (2021-09-15) Pix2seq (ViT-L) — 50.0 (2021-09-22) Pix2seq (R50-C4) — 47.3 (2021-09-22) Pix2seq (ViT-B) — 47.1 (2021-09-22) Pix2seq (R101-DC5) — 45.0 (2021-09-22) Pix2seq (R50-DC5 ) — 43.2 (2021-09-22) Pix2seq (R50) — 42.6 (2021-09-22) Pix2seq (ViT-L) — 50.0 (2021-09-22) Pix2seq (R50-C4) — 47.3 (2021-09-22) Pix2seq (ViT-B) — 47.1 (2021-09-22) Pix2seq (R101-DC5) — 45.0 (2021-09-22) Pix2seq (R50-DC5 ) — 43.2 (2021-09-22) Pix2seq (R50) — 42.6 (2021-09-22) Pix2seq (ViT-L) — 50.0 (2021-09-22) Pix2seq (R50-C4) — 47.3 (2021-09-22) Pix2seq (ViT-B) — 47.1 (2021-09-22) Pix2seq (R101-DC5) — 45.0 (2021-09-22) Pix2seq (R50-DC5 ) — 43.2 (2021-09-22) Pix2seq (R50) — 42.6 (2021-09-22) Pix2seq (ViT-L) — 50.0 (2021-09-22) Pix2seq (R50-C4) — 47.3 (2021-09-22) Pix2seq (ViT-B) — 47.1 (2021-09-22) Pix2seq (R101-DC5) — 45.0 (2021-09-22) Pix2seq (R50-DC5 ) — 43.2 (2021-09-22) Pix2seq (R50) — 42.6 (2021-09-22) Pix2seq (ViT-L) — 50.0 (2021-09-22) Pix2seq (R50-C4) — 47.3 (2021-09-22) Pix2seq (ViT-B) — 47.1 (2021-09-22) Pix2seq (R101-DC5) — 45.0 (2021-09-22) Pix2seq (R50-DC5 ) — 43.2 (2021-09-22) Pix2seq (R50) — 42.6 (2021-09-22) MAE (ViT-L, Mask R-CNN) — 53.3 (2021-11-11) MAE (ViT-B, Mask R-CNN) — 50.3 (2021-11-11) MAE (ViT-L, Mask R-CNN) — 53.3 (2021-11-11) MAE (ViT-B, Mask R-CNN) — 50.3 (2021-11-11) MAE (ViT-L, Mask R-CNN) — 53.3 (2021-11-11) MAE (ViT-B, Mask R-CNN) — 50.3 (2021-11-11) MAE (ViT-L, Mask R-CNN) — 53.3 (2021-11-11) MAE (ViT-B, Mask R-CNN) — 50.3 (2021-11-11) MAE (ViT-L, Mask R-CNN) — 53.3 (2021-11-11) MAE (ViT-B, Mask R-CNN) — 50.3 (2021-11-11) SwinV2-G (HTC++) — 62.5 (2021-11-18) SwinV2-G (HTC++) — 62.5 (2021-11-18) SwinV2-G (HTC++) — 62.5 (2021-11-18) SwinV2-G (HTC++) — 62.5 (2021-11-18) SwinV2-G (HTC++) — 62.5 (2021-11-18) Florence-CoSwin-H — 62.0 (2021-11-22) PoolFormer-S36 (Mask R-CNN) — 41.0 (2021-11-22) Florence-CoSwin-H — 62.0 (2021-11-22) PoolFormer-S36 (Mask R-CNN) — 41.0 (2021-11-22) Florence-CoSwin-H — 62.0 (2021-11-22) PoolFormer-S36 (Mask R-CNN) — 41.0 (2021-11-22) Florence-CoSwin-H — 62.0 (2021-11-22) PoolFormer-S36 (Mask R-CNN) — 41.0 (2021-11-22) Florence-CoSwin-H — 62.0 (2021-11-22) PoolFormer-S36 (Mask R-CNN) — 41.0 (2021-11-22) MAE-Det(MAE-Det-L+GFLV2) — 47.8 (2021-11-26) MAE-Det(MAE-Det-L+GFLV2) — 47.8 (2021-11-26) MAE-Det(MAE-Det-L+GFLV2) — 47.8 (2021-11-26) MAE-Det(MAE-Det-L+GFLV2) — 47.8 (2021-11-26) MAE-Det(MAE-Det-L+GFLV2) — 47.8 (2021-11-26) MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) — 58.7 (2021-12-02) MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) — 56.1 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, single-scale) — 54.3 (2021-12-02) MViT-L (Mask R-CNN, single-scale, IN21k pre-train) — 52.7 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) — 58.7 (2021-12-02) MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) — 56.1 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, single-scale) — 54.3 (2021-12-02) MViT-L (Mask R-CNN, single-scale, IN21k pre-train) — 52.7 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) — 58.7 (2021-12-02) MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) — 56.1 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, single-scale) — 54.3 (2021-12-02) MViT-L (Mask R-CNN, single-scale, IN21k pre-train) — 52.7 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) — 58.7 (2021-12-02) MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) — 56.1 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, single-scale) — 54.3 (2021-12-02) MViT-L (Mask R-CNN, single-scale, IN21k pre-train) — 52.7 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) — 58.7 (2021-12-02) MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) — 56.1 (2021-12-02) MViTv2-L (Cascade Mask R-CNN, single-scale) — 54.3 (2021-12-02) MViT-L (Mask R-CNN, single-scale, IN21k pre-train) — 52.7 (2021-12-02) GLIP (Swin-L, multi-scale) — 60.8 (2021-12-07) GLIP (Swin-L, multi-scale) — 60.8 (2021-12-07) GLIP (Swin-L, multi-scale) — 60.8 (2021-12-07) GLIP (Swin-L, multi-scale) — 60.8 (2021-12-07) GLIP (Swin-L, multi-scale) — 60.8 (2021-12-07) REGO-Deformable DETR-X101 — 49.1 (2021-12-09) REGO-Deformable DETR-X101 — 49.1 (2021-12-09) REGO-Deformable DETR-X101 — 49.1 (2021-12-09) REGO-Deformable DETR-X101 — 49.1 (2021-12-09) REGO-Deformable DETR-X101 — 49.1 (2021-12-09) ELSA-S (Cascade Mask RCNN) — 51.6 (2021-12-23) ELSA-S (Mask RCNN) — 48.3 (2021-12-23) ELSA-S (Cascade Mask RCNN) — 51.6 (2021-12-23) ELSA-S (Mask RCNN) — 48.3 (2021-12-23) ELSA-S (Cascade Mask RCNN) — 51.6 (2021-12-23) ELSA-S (Mask RCNN) — 48.3 (2021-12-23) ELSA-S (Cascade Mask RCNN) — 51.6 (2021-12-23) ELSA-S (Mask RCNN) — 48.3 (2021-12-23) ELSA-S (Cascade Mask RCNN) — 51.6 (2021-12-23) ELSA-S (Mask RCNN) — 48.3 (2021-12-23) PatchConvNet-S120 (Mask R-CNN) — 47.0 (2021-12-27) PatchConvNet-S60 (Mask R-CNN) — 46.4 (2021-12-27) PatchConvNet-S120 (Mask R-CNN) — 47.0 (2021-12-27) PatchConvNet-S60 (Mask R-CNN) — 46.4 (2021-12-27) PatchConvNet-S120 (Mask R-CNN) — 47.0 (2021-12-27) PatchConvNet-S60 (Mask R-CNN) — 46.4 (2021-12-27) PatchConvNet-S120 (Mask R-CNN) — 47.0 (2021-12-27) PatchConvNet-S60 (Mask R-CNN) — 46.4 (2021-12-27) PatchConvNet-S120 (Mask R-CNN) — 47.0 (2021-12-27) PatchConvNet-S60 (Mask R-CNN) — 46.4 (2021-12-27) DAB-DETR-DC5-R101 — 46.6 (2022-01-28) DAB-DETR-R101 — 44.1 (2022-01-28) DAB-DETR-DC5-R101 — 46.6 (2022-01-28) DAB-DETR-R101 — 44.1 (2022-01-28) DAB-DETR-DC5-R101 — 46.6 (2022-01-28) DAB-DETR-R101 — 44.1 (2022-01-28) DAB-DETR-DC5-R101 — 46.6 (2022-01-28) DAB-DETR-R101 — 44.1 (2022-01-28) DAB-DETR-DC5-R101 — 46.6 (2022-01-28) DAB-DETR-R101 — 44.1 (2022-01-28) CAE (ViT-L, Mask R-CNN, 1x schedule) — 54.5 (2022-02-07) CAE (ViT-L, Mask R-CNN, 1x schedule) — 54.5 (2022-02-07) CAE (ViT-L, Mask R-CNN, 1x schedule) — 54.5 (2022-02-07) CAE (ViT-L, Mask R-CNN, 1x schedule) — 54.5 (2022-02-07) CAE (ViT-L, Mask R-CNN, 1x schedule) — 54.5 (2022-02-07) DN-Deformable-DETR-R50++ — 49.5 (2022-03-02) DN-Deformable-DETR-R50++ — 49.5 (2022-03-02) DN-Deformable-DETR-R50++ — 49.5 (2022-03-02) DN-Deformable-DETR-R50++ — 49.5 (2022-03-02) DN-Deformable-DETR-R50++ — 49.5 (2022-03-02) DINO (Swin-L) — 63.2 (2022-03-07) DINO-5scale (24 epoch) — 51.3 (2022-03-07) DINO-5scale (36 epoch) — 51.2 (2022-03-07) DINO (Swin-L) — 63.2 (2022-03-07) DINO-5scale (24 epoch) — 51.3 (2022-03-07) DINO-5scale (36 epoch) — 51.2 (2022-03-07) DINO (Swin-L) — 63.2 (2022-03-07) DINO-5scale (24 epoch) — 51.3 (2022-03-07) DINO-5scale (36 epoch) — 51.2 (2022-03-07) DINO (Swin-L) — 63.2 (2022-03-07) DINO-5scale (24 epoch) — 51.3 (2022-03-07) DINO-5scale (36 epoch) — 51.2 (2022-03-07) DINO (Swin-L) — 63.2 (2022-03-07) DINO-5scale (24 epoch) — 51.3 (2022-03-07) DINO-5scale (36 epoch) — 51.2 (2022-03-07) ActiveMLP-B (Cascade Mask R-CNN) — 52.3 (2022-03-11) ActiveMLP-B (Cascade Mask R-CNN) — 52.3 (2022-03-11) ActiveMLP-B (Cascade Mask R-CNN) — 52.3 (2022-03-11) ActiveMLP-B (Cascade Mask R-CNN) — 52.3 (2022-03-11) ActiveMLP-B (Cascade Mask R-CNN) — 52.3 (2022-03-11) FocalNet-H (DINO) — 64.2 (2022-03-22) FocalNet-T (LRF, Cascade Mask R-CNN) — 51.5 (2022-03-22) FocalNet-H (DINO) — 64.2 (2022-03-22) FocalNet-T (LRF, Cascade Mask R-CNN) — 51.5 (2022-03-22) FocalNet-H (DINO) — 64.2 (2022-03-22) FocalNet-T (LRF, Cascade Mask R-CNN) — 51.5 (2022-03-22) FocalNet-H (DINO) — 64.2 (2022-03-22) FocalNet-T (LRF, Cascade Mask R-CNN) — 51.5 (2022-03-22) FocalNet-H (DINO) — 64.2 (2022-03-22) FocalNet-T (LRF, Cascade Mask R-CNN) — 51.5 (2022-03-22) ViTDet, ViT-H Cascade (multiscale) — 61.3 (2022-03-30) ViTDet, ViT-H Cascade — 60.4 (2022-03-30) ViTDet, ViT-H Cascade (multiscale) — 61.3 (2022-03-30) ViTDet, ViT-H Cascade — 60.4 (2022-03-30) ViTDet, ViT-H Cascade (multiscale) — 61.3 (2022-03-30) ViTDet, ViT-H Cascade — 60.4 (2022-03-30) ViTDet, ViT-H Cascade (multiscale) — 61.3 (2022-03-30) ViTDet, ViT-H Cascade — 60.4 (2022-03-30) ViTDet, ViT-H Cascade (multiscale) — 61.3 (2022-03-30) ViTDet, ViT-H Cascade — 60.4 (2022-03-30) DaViT-T (Mask R-CNN, 36 epochs) — 49.9 (2022-04-07) DaViT-T (Mask R-CNN, 36 epochs) — 49.9 (2022-04-07) DaViT-T (Mask R-CNN, 36 epochs) — 49.9 (2022-04-07) DaViT-T (Mask R-CNN, 36 epochs) — 49.9 (2022-04-07) DaViT-T (Mask R-CNN, 36 epochs) — 49.9 (2022-04-07) FAN-L-Hybrid — 55.1 (2022-04-26) FAN-L-Hybrid — 55.1 (2022-04-26) FAN-L-Hybrid — 55.1 (2022-04-26) FAN-L-Hybrid — 55.1 (2022-04-26) FAN-L-Hybrid — 55.1 (2022-04-26) ViT-Adapter-L (HTC++, BEiTv2 pretrain, multi-scale) — 60.5 (2022-05-17) ViT-Adapter-L (HTC++, BEiT pretrain, multi-scale) — 60.2 (2022-05-17) ViT-Adapter-L (HTC++, BEiTv2 pretrain, multi-scale) — 60.5 (2022-05-17) ViT-Adapter-L (HTC++, BEiT pretrain, multi-scale) — 60.2 (2022-05-17) ViT-Adapter-L (HTC++, BEiTv2 pretrain, multi-scale) — 60.5 (2022-05-17) ViT-Adapter-L (HTC++, BEiT pretrain, multi-scale) — 60.2 (2022-05-17) ViT-Adapter-L (HTC++, BEiTv2 pretrain, multi-scale) — 60.5 (2022-05-17) ViT-Adapter-L (HTC++, BEiT pretrain, multi-scale) — 60.2 (2022-05-17) ViT-Adapter-L (HTC++, BEiTv2 pretrain, multi-scale) — 60.5 (2022-05-17) ViT-Adapter-L (HTC++, BEiT pretrain, multi-scale) — 60.2 (2022-05-17) UM-MAE(HTC++, Swin-L, IN1K) — 57.4 (2022-05-20) UM-MAE(HTC++, Swin-L, IN1K) — 57.4 (2022-05-20) UM-MAE(HTC++, Swin-L, IN1K) — 57.4 (2022-05-20) UM-MAE(HTC++, Swin-L, IN1K) — 57.4 (2022-05-20) UM-MAE(HTC++, Swin-L, IN1K) — 57.4 (2022-05-20) HorNet-L — 59.2 (2022-07-28) HorNet-L — 59.2 (2022-07-28) HorNet-L — 59.2 (2022-07-28) HorNet-L — 59.2 (2022-07-28) HorNet-L — 59.2 (2022-07-28) MOAT-3 (IN-22K pretraining, single-scale) — 59.2 (2022-10-04) MOAT-2 (IN-22K pretraining, single-scale) — 58.5 (2022-10-04) MOAT-1 (IN-1K pretraining, single-scale) — 57.7 (2022-10-04) MOAT-0 (IN-1K pretraining, single-scale) — 55.9 (2022-10-04) tiny-MOAT-3 (IN-1K pretraining, single-scale) — 55.2 (2022-10-04) tiny-MOAT-2 (IN-1K pretraining, single-scale) — 53.0 (2022-10-04) tiny-MOAT-1 (IN-1K pretraining, single-scale) — 51.9 (2022-10-04) tiny-MOAT-0 (IN-1K pretraining, single-scale) — 50.5 (2022-10-04) MOAT-3 (IN-22K pretraining, single-scale) — 59.2 (2022-10-04) MOAT-2 (IN-22K pretraining, single-scale) — 58.5 (2022-10-04) MOAT-1 (IN-1K pretraining, single-scale) — 57.7 (2022-10-04) MOAT-0 (IN-1K pretraining, single-scale) — 55.9 (2022-10-04) tiny-MOAT-3 (IN-1K pretraining, single-scale) — 55.2 (2022-10-04) tiny-MOAT-2 (IN-1K pretraining, single-scale) — 53.0 (2022-10-04) tiny-MOAT-1 (IN-1K pretraining, single-scale) — 51.9 (2022-10-04) tiny-MOAT-0 (IN-1K pretraining, single-scale) — 50.5 (2022-10-04) MOAT-3 (IN-22K pretraining, single-scale) — 59.2 (2022-10-04) MOAT-2 (IN-22K pretraining, single-scale) — 58.5 (2022-10-04) MOAT-1 (IN-1K pretraining, single-scale) — 57.7 (2022-10-04) MOAT-0 (IN-1K pretraining, single-scale) — 55.9 (2022-10-04) tiny-MOAT-3 (IN-1K pretraining, single-scale) — 55.2 (2022-10-04) tiny-MOAT-2 (IN-1K pretraining, single-scale) — 53.0 (2022-10-04) tiny-MOAT-1 (IN-1K pretraining, single-scale) — 51.9 (2022-10-04) tiny-MOAT-0 (IN-1K pretraining, single-scale) — 50.5 (2022-10-04) MOAT-3 (IN-22K pretraining, single-scale) — 59.2 (2022-10-04) MOAT-2 (IN-22K pretraining, single-scale) — 58.5 (2022-10-04) MOAT-1 (IN-1K pretraining, single-scale) — 57.7 (2022-10-04) MOAT-0 (IN-1K pretraining, single-scale) — 55.9 (2022-10-04) tiny-MOAT-3 (IN-1K pretraining, single-scale) — 55.2 (2022-10-04) tiny-MOAT-2 (IN-1K pretraining, single-scale) — 53.0 (2022-10-04) tiny-MOAT-1 (IN-1K pretraining, single-scale) — 51.9 (2022-10-04) tiny-MOAT-0 (IN-1K pretraining, single-scale) — 50.5 (2022-10-04) MOAT-3 (IN-22K pretraining, single-scale) — 59.2 (2022-10-04) MOAT-2 (IN-22K pretraining, single-scale) — 58.5 (2022-10-04) MOAT-1 (IN-1K pretraining, single-scale) — 57.7 (2022-10-04) MOAT-0 (IN-1K pretraining, single-scale) — 55.9 (2022-10-04) tiny-MOAT-3 (IN-1K pretraining, single-scale) — 55.2 (2022-10-04) tiny-MOAT-2 (IN-1K pretraining, single-scale) — 53.0 (2022-10-04) tiny-MOAT-1 (IN-1K pretraining, single-scale) — 51.9 (2022-10-04) tiny-MOAT-0 (IN-1K pretraining, single-scale) — 50.5 (2022-10-04) TEC(VIT-B, Mask-RCNN) — 54.6 (2022-10-20) TEC(VIT-B, Mask-RCNN) — 54.6 (2022-10-20) TEC(VIT-B, Mask-RCNN) — 54.6 (2022-10-20) TEC(VIT-B, Mask-RCNN) — 54.6 (2022-10-20) TEC(VIT-B, Mask-RCNN) — 54.6 (2022-10-20) Frozen Backbone, SwinV2-G-ext22K (HTC) — 59.3 (2022-11-03) Frozen Backbone, SwinV2-G-ext22K (HTC) — 59.3 (2022-11-03) Frozen Backbone, SwinV2-G-ext22K (HTC) — 59.3 (2022-11-03) Frozen Backbone, SwinV2-G-ext22K (HTC) — 59.3 (2022-11-03) Frozen Backbone, SwinV2-G-ext22K (HTC) — 59.3 (2022-11-03) InternImage-H — 65.0 (2022-11-10) InternImage-XL — 64.2 (2022-11-10) InternImage-H — 65.0 (2022-11-10) InternImage-XL — 64.2 (2022-11-10) InternImage-H — 65.0 (2022-11-10) InternImage-XL — 64.2 (2022-11-10) InternImage-H — 65.0 (2022-11-10) InternImage-XL — 64.2 (2022-11-10) InternImage-H — 65.0 (2022-11-10) InternImage-XL — 64.2 (2022-11-10) EVA — 64.5 (2022-11-14) EVA — 64.5 (2022-11-14) EVA — 64.5 (2022-11-14) EVA — 64.5 (2022-11-14) EVA — 64.5 (2022-11-14) M3I Pre-training (InternImage-H) — 65.0 (2022-11-17) M3I Pre-training (InternImage-H) — 65.0 (2022-11-17) M3I Pre-training (InternImage-H) — 65.0 (2022-11-17) M3I Pre-training (InternImage-H) — 65.0 (2022-11-17) M3I Pre-training (InternImage-H) — 65.0 (2022-11-17) Co-DETR — 65.9 (2022-11-22) Co-DETR (Swin-L) — 64.7 (2022-11-22) Co-DETR — 65.9 (2022-11-22) Co-DETR (Swin-L) — 64.7 (2022-11-22) Co-DETR — 65.9 (2022-11-22) Co-DETR (Swin-L) — 64.7 (2022-11-22) Co-DETR — 65.9 (2022-11-22) Co-DETR (Swin-L) — 64.7 (2022-11-22) Co-DETR — 65.9 (2022-11-22) Co-DETR (Swin-L) — 64.7 (2022-11-22) RevCol-H(DINO) — 63.8 (2022-12-22) RevCol-H(DINO) — 63.8 (2022-12-22) RevCol-H(DINO) — 63.8 (2022-12-22) RevCol-H(DINO) — 63.8 (2022-12-22) RevCol-H(DINO) — 63.8 (2022-12-22) YOLOv6-L6(46 fps, 1280, V100) — 57.2 (2023-01-13) YOLOv6-L6(46 fps, 1280, V100) — 57.2 (2023-01-13) YOLOv6-L6(46 fps, 1280, V100) — 57.2 (2023-01-13) YOLOv6-L6(46 fps, 1280, V100) — 57.2 (2023-01-13) YOLOv6-L6(46 fps, 1280, V100) — 57.2 (2023-01-13) Grounding DINO — 63.0 (2023-03-09) Grounding DINO — 63.0 (2023-03-09) Grounding DINO — 63.0 (2023-03-09) Grounding DINO — 63.0 (2023-03-09) Grounding DINO — 63.0 (2023-03-09) UNINEXT-H — 60.6 (2023-03-12) UNINEXT-H — 60.6 (2023-03-12) UNINEXT-H — 60.6 (2023-03-12) UNINEXT-H — 60.6 (2023-03-12) UNINEXT-H — 60.6 (2023-03-12) Focal-Stable-DINO (Focal-Huge, no TTA) — 64.6 (2023-04-25) Focal-Stable-DINO (Focal-Huge, no TTA) — 64.6 (2023-04-25) Focal-Stable-DINO (Focal-Huge, no TTA) — 64.6 (2023-04-25) Focal-Stable-DINO (Focal-Huge, no TTA) — 64.6 (2023-04-25) Focal-Stable-DINO (Focal-Huge, no TTA) — 64.6 (2023-04-25) Hiera-L — 55.0 (2023-06-01) Hiera-L — 55.0 (2023-06-01) Hiera-L — 55.0 (2023-06-01) Hiera-L — 55.0 (2023-06-01) Hiera-L — 55.0 (2023-06-01) TransNeXt-Base (IN-1K pretrain, DINO 1x) — 57.1 (2023-11-28) TransNeXt-Small (IN-1K pretrain, DINO 1x) — 56.6 (2023-11-28) TransNeXt-Tiny (IN-1K pretrain, DINO 1x) — 55.7 (2023-11-28) TransNeXt-Base (IN-1K pretrain, DINO 1x) — 57.1 (2023-11-28) TransNeXt-Small (IN-1K pretrain, DINO 1x) — 56.6 (2023-11-28) TransNeXt-Tiny (IN-1K pretrain, DINO 1x) — 55.7 (2023-11-28) TransNeXt-Base (IN-1K pretrain, DINO 1x) — 57.1 (2023-11-28) TransNeXt-Small (IN-1K pretrain, DINO 1x) — 56.6 (2023-11-28) TransNeXt-Tiny (IN-1K pretrain, DINO 1x) — 55.7 (2023-11-28) TransNeXt-Base (IN-1K pretrain, DINO 1x) — 57.1 (2023-11-28) TransNeXt-Small (IN-1K pretrain, DINO 1x) — 56.6 (2023-11-28) TransNeXt-Tiny (IN-1K pretrain, DINO 1x) — 55.7 (2023-11-28) TransNeXt-Base (IN-1K pretrain, DINO 1x) — 57.1 (2023-11-28) TransNeXt-Small (IN-1K pretrain, DINO 1x) — 56.6 (2023-11-28) TransNeXt-Tiny (IN-1K pretrain, DINO 1x) — 55.7 (2023-11-28) GLEE-Pro — 62.0 (2023-12-14) GLEE-Plus — 60.4 (2023-12-14) GLEE-Lite — 55.0 (2023-12-14) GLEE-Pro — 62.0 (2023-12-14) GLEE-Plus — 60.4 (2023-12-14) GLEE-Lite — 55.0 (2023-12-14) GLEE-Pro — 62.0 (2023-12-14) GLEE-Plus — 60.4 (2023-12-14) GLEE-Lite — 55.0 (2023-12-14) GLEE-Pro — 62.0 (2023-12-14) GLEE-Plus — 60.4 (2023-12-14) GLEE-Lite — 55.0 (2023-12-14) GLEE-Pro — 62.0 (2023-12-14) GLEE-Plus — 60.4 (2023-12-14) GLEE-Lite — 55.0 (2023-12-14) ViT-CoMer — 64.3 (2024-03-13) ViT-CoMer — 64.3 (2024-03-13) ViT-CoMer — 64.3 (2024-03-13) ViT-CoMer — 64.3 (2024-03-13) ViT-CoMer — 64.3 (2024-03-13) CP-DETR-L Swin-L(Fine tuning separately in COCO) — 64.1 (2024-12-13) CP-DETR-L Swin-L(Fine tuning separately in COCO) — 64.1 (2024-12-13) CP-DETR-L Swin-L(Fine tuning separately in COCO) — 64.1 (2024-12-13) CP-DETR-L Swin-L(Fine tuning separately in COCO) — 64.1 (2024-12-13) CP-DETR-L Swin-L(Fine tuning separately in COCO) — 64.1 (2024-12-13) PE_spatial (DETA) — 66.0 (2025-04-17) PE_spatial (DETA) — 66.0 (2025-04-17) PE_spatial (DETA) — 66.0 (2025-04-17) PE_spatial (DETA) — 66.0 (2025-04-17) PE_spatial (DETA) — 66.0 (2025-04-17) Cascade Mask R-CNN (ResNet-50) — 46.3 (2015-12-10) Mask R-CNN (ResNeXt-152-FPN, cascade) — 48.6 (2018-11-21) EfficientDet-D7 (1536) — 52.1 (2019-11-20) RetinaNet (SpineNet-190, 1536x1536) — 52.2 (2019-12-10) ResNeSt-200 (multi-scale) — 52.47 (2020-04-19) SpineNet-190 (1280, with Self-training on OpenImages, single-scale) — 54.2 (2020-06-11) YOLOv4-P7 CSP-P7 (single-scale, 16 fps) — 55.4 (2020-11-16) Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) — 57.0 (2020-12-13) Swin-L (HTC++, multi scale) — 58.0 (2021-03-25) DyHead (Swin-L, multi scale, self-training) — 60.3 (2021-06-15) Soft Teacher + Swin-L (HTC++, multi-scale) — 60.7 (2021-06-16) SwinV2-G (HTC++) — 62.5 (2021-11-18) DINO (Swin-L) — 63.2 (2022-03-07) FocalNet-H (DINO) — 64.2 (2022-03-22) InternImage-H — 65.0 (2022-11-10) Co-DETR — 65.9 (2022-11-22) PE_spatial (DETA) — 66.0 (2025-04-17)
RankModel box APAP50AP75APSAPMAPLParams (M) Extra Training Data PaperCodeYear
1 PE_spatial (DETA) 66.0–––––1900 ✓ Perception Encoder: The best visual embeddings are not at the output of the network facebookresearch/perception_models 2025
2 Co-DETR 65.9–––––314 ✓ DETRs with Collaborative Hybrid Assignments Training open-mmlab/mmdetection · siyuanliii/masa · sense-x/co-detr · +3 2022
3 M3I Pre-training (InternImage-H) 65.0–––––– ✓ Towards All-in-one Pre-training via Maximizing Multi-modal Mutual Information OpenGVLab/M3I-Pretraining 2022
3 InternImage-H 65.0–––––– ✓ InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions opengvlab/internimage · OpenGVLab/M3I-Pretraining · chenller/mmseg-extension 2022
5 Co-DETR (Swin-L) 64.7–––––218 ✓ DETRs with Collaborative Hybrid Assignments Training open-mmlab/mmdetection · siyuanliii/masa · sense-x/co-detr · +3 2022
6 Focal-Stable-DINO (Focal-Huge, no TTA) 64.681.571.450.468.578.5689 ✓ A Strong and Reproducible Object Detector with Only Public Datasets microsoft/FocalNet · idea-research/stable-dino · idea-research/stabledino 2023
7 EVA 64.582.170.849.468.478.5– ✓ EVA: Exploring the Limits of Masked Visual Representation Learning at Scale rwightman/pytorch-image-models · open-mmlab/mmselfsup · baaivision/eva · +3 2022
8 ViT-CoMer 64.3–––––363 ViT-CoMer: Vision Transformer with Convolutional Multi-scale Feature Interaction for Dense Predictions Traffic-X/ViT-CoMer · chenller/mmseg-extension 2024
9 FocalNet-H (DINO) 64.2–––––– ✓ Focal Modulation Networks PaddlePaddle/PaddleDetection · keras-team/keras-io · microsoft/FocalNet · +6 2022
9 InternImage-XL 64.2–––––– ✓ InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions opengvlab/internimage · OpenGVLab/M3I-Pretraining · chenller/mmseg-extension 2022
11 CP-DETR-L Swin-L(Fine tuning separately in COCO) 64.1–––––– ✓ CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection 2024
12 RevCol-H(DINO) 63.8–––––– ✓ Reversible Column Networks megvii-research/revcol 2022
13 DINO (Swin-L) 63.2–––––– DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection IDEA-Research/Grounded-Segment-Anything · PaddlePaddle/PaddleDetection · lucasjinreal/yolov7_d2 · +13 2022
14 Grounding DINO 63.0–––––– ✓ Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection huggingface/transformers · IDEA-Research/Grounded-Segment-Anything · idea-research/groundingdino · +7 2023
15 SwinV2-G (HTC++) 62.5–––––– ✓ Swin Transformer V2: Scaling Up Capacity and Resolution rwightman/pytorch-image-models · microsoft/Swin-Transformer · PaddlePaddle/PaddleDetection · +20 2021
16 Florence-CoSwin-H 62–––––– ✓ Florence: A New Foundation Model for Computer Vision microsoft/unicl · MindCode-4/code-3 2021
16 GLEE-Pro 62.0–––––– ✓ General Object Foundation Model for Images and Videos at Scale FoundationVision/GLEE 2023
18 ViTDet, ViT-H Cascade (multiscale) 61.3–––––– Exploring Plain Vision Transformer Backbones for Object Detection facebookresearch/detectron2 · PaddlePaddle/PaddleDetection · alibaba/EasyCV · +8 2022
19 GLIP (Swin-L, multi-scale) 60.8–––––– ✓ Grounded Language-Image Pre-training microsoft/GLIP · brown-palm/ObjectPrompt · rsCPSyEu/ovd_cod 2021
20 Soft Teacher + Swin-L (HTC++, multi-scale) 60.7–––––– ✓ End-to-End Semi-Supervised Object Detection with Soft Teacher microsoft/SoftTeacher · amazon-science/bigdetection · amazon-research/bigdetection · +5 2021
21 UNINEXT-H 60.677.566.745.164.875.3– ✓ Universal Instance Perception as Object Discovery and Retrieval MasterBin-IIAU/UNINEXT 2023
22 ViT-Adapter-L (HTC++, BEiTv2 pretrain, multi-scale) 60.5–––––– Vision Transformer Adapter for Dense Predictions czczup/vit-adapter · chenller/mmseg-extension 2022
23 ViTDet, ViT-H Cascade 60.4–––––– Exploring Plain Vision Transformer Backbones for Object Detection facebookresearch/detectron2 · PaddlePaddle/PaddleDetection · alibaba/EasyCV · +8 2022
23 GLEE-Plus 60.4–––––– ✓ General Object Foundation Model for Images and Videos at Scale FoundationVision/GLEE 2023
25 DyHead (Swin-L, multi scale, self-training) 60.378.2–––74.2– ✓ Dynamic Head: Unifying Object Detection Heads with Attentions open-mmlab/mmdetection · microsoft/DynamicHead · Coldestadam/DynamicHead 2021
26 ViT-Adapter-L (HTC++, BEiT pretrain, multi-scale) 60.2–––––– Vision Transformer Adapter for Dense Predictions czczup/vit-adapter · chenller/mmseg-extension 2022
27 Soft Teacher+Swin-L(HTC++, single scale) 60.1–––––– ✓ End-to-End Semi-Supervised Object Detection with Soft Teacher microsoft/SoftTeacher · amazon-science/bigdetection · amazon-research/bigdetection · +5 2021
28 CBNetV2 (Dual-Swin-L HTC, multi-scale) 59.6–––––– CBNet: A Composite Backbone Network Architecture for Object Detection PaddlePaddle/PaddleDetection · shinya7y/UniverseNet · VDIGPKU/CBNetV2 · +1 2021
29 Frozen Backbone, SwinV2-G-ext22K (HTC) 59.3–––––– Could Giant Pretrained Image Models Extract Universal Representations? 2022
30 HorNet-L 59.2–––––– HorNet: Efficient High-Order Spatial Interactions with Recursive Gated Convolutions open-mmlab/mmclassification · towhee-io/towhee · chengtan9907/OpenSTL · +5 2022
30 MOAT-3 (IN-22K pretraining, single-scale) 59.2–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
32 CBNetV2 (Dual-Swin-L HTC, multi-scale) 59.1–––––– CBNet: A Composite Backbone Network Architecture for Object Detection PaddlePaddle/PaddleDetection · shinya7y/UniverseNet · VDIGPKU/CBNetV2 · +1 2021
33 Focal-L (DyHead, multi-scale) 58.777.2–––73.4– Focal Self-attention for Local-Global Interactions in Vision Transformers BR-IDL/PaddleViT · microsoft/Focal-Transformer · microsoft/esvit 2021
33 MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) 58.7–––––– MViTv2: Improved Multiscale Vision Transformers for Classification and Detection rwightman/pytorch-image-models · facebookresearch/detectron2 · facebookresearch/SlowFast · +6 2021
35 MOAT-2 (IN-22K pretraining, single-scale) 58.5–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
36 DyHead (Swin-L, multi scale) 58.476.8–44.562.273.2– Dynamic Head: Unifying Object Detection Heads with Attentions open-mmlab/mmdetection · microsoft/DynamicHead · Coldestadam/DynamicHead 2021
37 Swin-L (HTC++, multi scale) 58–––––– Swin Transformer: Hierarchical Vision Transformer using Shifted Windows huggingface/transformers · rwightman/pytorch-image-models · open-mmlab/mmdetection · +77 2021
38 MOAT-1 (IN-1K pretraining, single-scale) 57.7–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
39 UM-MAE(HTC++, Swin-L, IN1K) 57.4–––––– Uniform Masking: Enabling MAE Pre-training for Pyramid-based Vision Transformers with Locality implus/um-mae 2022
40 YOLOv6-L6(46 fps, 1280, V100) 57.274.5––––– YOLOv6 v3.0: A Full-Scale Reloading PaddlePaddle/PaddleDetection · meituan/yolov6 · PaddlePaddle/PaddleYOLO · +2 2023
41 Swin-L (HTC++, single scale) 57.1–––––– Swin Transformer: Hierarchical Vision Transformer using Shifted Windows huggingface/transformers · rwightman/pytorch-image-models · open-mmlab/mmdetection · +77 2021
41 TransNeXt-Base (IN-1K pretrain, DINO 1x) 57.1–––––– TransNeXt: Robust Foveal Visual Perception for Vision Transformers Westlake-AI/openmixup · daishiresearch/transnext · chenller/mmseg-extension · +1 2023
43 Cascade Eff-B7 NAS-FPN (1280, self-training Copy Paste, single-scale) 57.0–––––– ✓ Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation PaddlePaddle/PaddleOCR · open-mmlab/mmdetection · tensorflow/tpu · +2 2020
44 TransNeXt-Small (IN-1K pretrain, DINO 1x) 56.6–––––– TransNeXt: Robust Foveal Visual Perception for Vision Transformers Westlake-AI/openmixup · daishiresearch/transnext · chenller/mmseg-extension · +1 2023
45 QueryInst (single scale) 56.175.861.740.259.871.5– Instances as Queries open-mmlab/mmdetection · hustvl/QueryInst · Bo396543018/picodet_repro · +2 2021
45 MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) 56.1–––––– MViTv2: Improved Multiscale Vision Transformers for Classification and Detection rwightman/pytorch-image-models · facebookresearch/detectron2 · facebookresearch/SlowFast · +6 2021
47 MOAT-0 (IN-1K pretraining, single-scale) 55.9–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
48 TransNeXt-Tiny (IN-1K pretrain, DINO 1x) 55.7–––––– TransNeXt: Robust Foveal Visual Perception for Vision Transformers Westlake-AI/openmixup · daishiresearch/transnext · chenller/mmseg-extension · +1 2023
49 YOLOv4-P7 CSP-P7 (single-scale, 16 fps) 55.473.360.738.159.567.4– Scaled-YOLOv4: Scaling Cross Stage Partial Network AlexeyAB/darknet · RangiLyu/nanodet · WongKinYiu/ScaledYOLOv4 · +38 2020
50 tiny-MOAT-3 (IN-1K pretraining, single-scale) 55.2–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
51 FAN-L-Hybrid 55.1–––––– Understanding The Robustness in Vision Transformers nvlabs/fan · NVlabs/STL 2022
52 Hiera-L 55–––––– Hiera: A Hierarchical Vision Transformer without the Bells-and-Whistles huggingface/pytorch-image-models · facebookresearch/hiera · leondgarse/keras_cv_attention_models · +1 2023
52 GLEE-Lite 55.0–––––– ✓ General Object Foundation Model for Images and Videos at Scale FoundationVision/GLEE 2023
54 TEC(VIT-B, Mask-RCNN) 54.6–––––– Towards Sustainable Self-supervised Learning sail-sg/tec 2022
55 Cascade Eff-B7 NAS-FPN (1280) 54.5–––––– Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation PaddlePaddle/PaddleOCR · open-mmlab/mmdetection · tensorflow/tpu · +2 2020
55 CAE (ViT-L, Mask R-CNN, 1x schedule) 54.5–––––– Context Autoencoder for Self-Supervised Representation Learning open-mmlab/mmselfsup · PaddlePaddle/PaddleFL · PaddlePaddle/VIMER · +3 2022
57 MViTv2-L (Cascade Mask R-CNN, single-scale) 54.3–––––– MViTv2: Improved Multiscale Vision Transformers for Classification and Detection rwightman/pytorch-image-models · facebookresearch/detectron2 · facebookresearch/SlowFast · +6 2021
58 SpineNet-190 (1280, with Self-training on OpenImages, single-scale) 54.2–––––– ✓ Rethinking Pre-training and Self-training tensorflow/tpu · stanleyjzheng/PyData 2020
59 Cascade RCNN-RS (SpineNet-143L, single scale) 53.6––34.556.770.6– Simple Training Strategies and Model Scaling for Object Detection tensorflow/tpu 2021
60 UniverseNet-20.08d (Res2Net-101, DCN, multi-scale) 53.570.858.936.957.568.1– USB: Universal-Scale Object Detection Benchmark shinya7y/UniverseNet 2021
61 MAE (ViT-L, Mask R-CNN) 53.3–––––– Masked Autoencoders Are Scalable Vision Learners facebookresearch/mae · lightly-ai/lightly · open-mmlab/mmselfsup · +55 2021
62 Cascade RCNN-RS (ResNet-200, single scale) 53.1––33.956.270.3– Simple Training Strategies and Model Scaling for Object Detection tensorflow/tpu 2021
63 tiny-MOAT-2 (IN-1K pretraining, single-scale) 53.0–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
64 MViT-L (Mask R-CNN, single-scale, IN21k pre-train) 52.7–––––– MViTv2: Improved Multiscale Vision Transformers for Classification and Detection rwightman/pytorch-image-models · facebookresearch/detectron2 · facebookresearch/SlowFast · +6 2021
65 ResNeSt-200 (multi-scale) 52.4771.0057.0736.8056.3666.29– ResNeSt: Split-Attention Networks rwightman/pytorch-image-models · open-mmlab/mmdetection · open-mmlab/mmpose · +33 2020
66 ActiveMLP-B (Cascade Mask R-CNN) 52.3–––––– Active Token Mixer microsoft/TokenMixers · microsoft/activemlp 2022
67 RetinaNet (SpineNet-190, 1536x1536) 52.2–––––– SpineNet: Learning Scale-Permuted Backbone for Recognition and Localization tensorflow/models · tensorflow/tpu · tensorflow/tpu · +10 2019
68 EfficientDet-D7 (1536) 52.1–––––– EfficientDet: Scalable and Efficient Object Detection tensorflow/models · PaddlePaddle/PaddleDetection · google/automl · +61 2019
69 tiny-MOAT-1 (IN-1K pretraining, single-scale) 51.9–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
70 GCNet (ResNeXt-101 + DCN + cascade + GC r4) 51.870.456.1–––– Global Context Networks rwightman/pytorch-image-models · PaddlePaddle/PaddleDetection · xvjiarui/GCNet 2020
71 ELSA-S (Cascade Mask RCNN) 51.670.556.0–––– ELSA: Enhanced Local Self-Attention for Vision Transformer damo-cv/elsa 2021
72 FocalNet-T (LRF, Cascade Mask R-CNN) 51.570.356.0–––– Focal Modulation Networks PaddlePaddle/PaddleDetection · keras-team/keras-io · microsoft/FocalNet · +6 2022
73 DINO-5scale (24 epoch) 51.369.15634.554.265.8– DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection IDEA-Research/Grounded-Segment-Anything · PaddlePaddle/PaddleDetection · lucasjinreal/yolov7_d2 · +13 2022
74 DINO-5scale (36 epoch) 51.26955.83554.365.3– DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection IDEA-Research/Grounded-Segment-Anything · PaddlePaddle/PaddleDetection · lucasjinreal/yolov7_d2 · +13 2022
75 ResNeSt-200-DCN (single-scale) 50.9169.5355.4032.6754.6665.83– ResNeSt: Split-Attention Networks rwightman/pytorch-image-models · open-mmlab/mmdetection · open-mmlab/mmpose · +33 2020
76 UniverseNet-20.08d (Res2Net-101, DCN, single-scale) 50.969.555.433.555.565.8– USB: Universal-Scale Object Detection Benchmark shinya7y/UniverseNet 2021
77 ResNeSt-200 (single-scale) 50.5468.7855.17–54.263.9– ResNeSt: Split-Attention Networks rwightman/pytorch-image-models · open-mmlab/mmdetection · open-mmlab/mmpose · +33 2020
78 tiny-MOAT-0 (IN-1K pretraining, single-scale) 50.5–––––– MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models google-research/deeplab2 · RooKichenn/pytorch-MOAT 2022
79 MAE (ViT-B, Mask R-CNN) 50.3–––––– Masked Autoencoders Are Scalable Vision Learners facebookresearch/mae · lightly-ai/lightly · open-mmlab/mmselfsup · +55 2021
80 Sparse R-CNN (PVTv2-B2) 50.169.554.9–––– PVT v2: Improved Baselines with Pyramid Vision Transformer rwightman/pytorch-image-models · open-mmlab/mmdetection · open-mmlab/mmpose · +15 2021
81 Pix2seq (ViT-L) 50.0–––––– ✓ Pix2seq: A Language Modeling Framework for Object Detection google-research/pix2seq · gaopengcuhk/Stable-Pix2Seq · gaopengcuhk/Unofficial-Pix2Seq · +3 2021
82 DaViT-T (Mask R-CNN, 36 epochs) 49.9–––––– DaViT: Dual Attention Vision Transformers rwightman/pytorch-image-models · leondgarse/keras_cv_attention_models · dingmyu/davit · +1 2022
83 BoTNet 200 (Mask R-CNN, single scale, 72 epochs) 49.771.354.6–––– Bottleneck Transformers for Visual Recognition rwightman/pytorch-image-models · BR-IDL/PaddleViT · The-AI-Summer/self_attention · +10 2021
84 BoTNet 152 (Mask R-CNN, single scale, 72 epochs) 49.57154.2–––– Bottleneck Transformers for Visual Recognition rwightman/pytorch-image-models · BR-IDL/PaddleViT · The-AI-Summer/self_attention · +10 2021
84 DN-Deformable-DETR-R50++ 49.567.653.831.352.665.447 DN-DETR: Accelerate DETR Training by Introducing Query DeNoising IDEACVR/DINO · idea-research/dino · IDEA-Research/detrex · +14 2022
86 REGO-Deformable DETR-X101 49.167.553.13052.665– Recurrent Glimpse-based Decoder for Detection with Transformer zhechen/deformable-detr-rego 2021
87 CenterMask+VoVNet99 (multi-scale) 48.667.8––––– CenterMask : Real-Time Anchor-Free Instance Segmentation youngwanLEE/centermask2 · youngwanLEE/CenterMask · youngwanLEE/vovnet-detectron2 · +5 2019
87 Mask R-CNN (ResNeXt-152-FPN, cascade) 48.666.852.9–––– Rethinking ImageNet Pre-training tensorpack/tensorpack 2018
89 UniverseNet-20.08 (Res2Net-50, DCN, single-scale) 48.567.052.630.652.762.7– USB: Universal-Scale Object Detection Benchmark shinya7y/UniverseNet 2021
89 XCiT-M24/8 48.5–––––– XCiT: Cross-Covariance Image Transformers rwightman/pytorch-image-models · facebookresearch/dino · facebookresearch/vissl · +9 2021
91 ELSA-S (Mask RCNN) 48.370.452.9–––– ELSA: Enhanced Local Self-Attention for Vision Transformer damo-cv/elsa 2021
92 XCiT-S24/8 48.1–––––– XCiT: Cross-Covariance Image Transformers rwightman/pytorch-image-models · facebookresearch/dino · facebookresearch/vissl · +9 2021
93 GCNet (ResNeXt-101 + DCN + cascade + GC r16) 47.966.952.2–––– GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond open-mmlab/mmdetection · open-mmlab/mmsegmentation · PaddlePaddle/PaddleSeg · +6 2019
94 MAE-Det(MAE-Det-L+GFLV2) 47.865.552.230.351.961.1– MAE-DET: Revisiting Maximum Entropy Principle in Zero-Shot NAS for Efficient Object Detection alibaba/lightweight-neural-architecture-search 2021
95 Res2Net101+HTC 47.566.551.328.651.662.1– Res2Net: A New Multi-scale Backbone Architecture rwightman/pytorch-image-models · open-mmlab/mmdetection · PaddlePaddle/PaddleDetection · +31 2019
96 Mask R-CNN (ResNet-101-FPN, GN, Cascade) 47.4–––––– Rethinking ImageNet Pre-training tensorpack/tensorpack 2018
97 Pix2seq (R50-C4) 47.3–––––– Pix2seq: A Language Modeling Framework for Object Detection google-research/pix2seq · gaopengcuhk/Stable-Pix2Seq · gaopengcuhk/Unofficial-Pix2Seq · +3 2021
98 Pix2seq (ViT-B) 47.1–––––– Pix2seq: A Language Modeling Framework for Object Detection google-research/pix2seq · gaopengcuhk/Stable-Pix2Seq · gaopengcuhk/Unofficial-Pix2Seq · +3 2021
99 HTC (HRNetV2p-W48) 47.0––28.850.362.2– Deep High-Resolution Representation Learning for Visual Recognition open-mmlab/mmdetection · PaddlePaddle/PaddleDetection · open-mmlab/mmsegmentation · +39 2019
99 PatchConvNet-S120 (Mask R-CNN) 47.0–––––– Augmenting Convolutional networks with attention-based aggregation facebookresearch/deit · keras-team/keras-io · DarshanDeshpande/jax-models · +2 2021
1–100 / 1100 다음 → 페이지당 10 20 50 100