| Rank | Model |
mIoU (val) | mIoU (test) |
Paper | Code | Year |
| 1 |
TEC (ViT-B/16, 224x224, SSL+FT, mmseg) |
63.2 | 62.5 |
Towards Sustainable Self-supervised Learning
|
sail-sg/tec |
2022 |
| 2 |
SERE (ViT-B/16, 100ep, 224x224, SSL+FT) |
63.0 | 63.3 |
SERE: Exploring Feature Self-relation for Self-supervised Transformer
|
MCG-NKU/SERE |
2022 |
| 3 |
TEC (ViT-B/16, 224x224, SSL+FT) |
62.0 | – |
Towards Sustainable Self-supervised Learning
|
sail-sg/tec |
2022 |
| 4 |
MAE (ViT-B/16, 224x224, SSL+FT, mmseg) |
61.6 | 61.2 |
Masked Autoencoders Are Scalable Vision Learners
|
facebookresearch/mae · lightly-ai/lightly · open-mmlab/mmselfsup
· +55 |
2021 |
| 5 |
MAE (ViT-B/16, 224x224, SSL+FT) |
61.0 | 60.2 |
Masked Autoencoders Are Scalable Vision Learners
|
facebookresearch/mae · lightly-ai/lightly · open-mmlab/mmselfsup
· +55 |
2021 |
| 6 |
SERE (ViT-S/16, 100ep, 224x224, SSL+FT, mmseg) |
59.4 | 59.0 |
SERE: Exploring Feature Self-relation for Self-supervised Transformer
|
MCG-NKU/SERE |
2022 |
| 7 |
SERE (ViT-S/16, 100ep, 224x224, SSL+FT) |
58.9 | 57.8 |
SERE: Exploring Feature Self-relation for Self-supervised Transformer
|
MCG-NKU/SERE |
2022 |
| 8 |
RF-ConvNext-Tiny (rfmerge, P4, 224x224, SUP) |
51.3 | 51.1 |
RF-Next: Efficient Receptive Field Search for Convolutional Neural Networks
|
ShangHua-Gao/G2L-search · ShangHua-Gao/RFNext |
2022 |
| 9 |
RF-ConvNext-Tiny (rfmultiple, P4, 224x224, SUP) |
50.8 | 50.5 |
RF-Next: Efficient Receptive Field Search for Convolutional Neural Networks
|
ShangHua-Gao/G2L-search · ShangHua-Gao/RFNext |
2022 |
| 10 |
RF-ConvNext-Tiny (rfsingle, P4, 224x224, SUP) |
50.7 | 50.5 |
RF-Next: Efficient Receptive Field Search for Convolutional Neural Networks
|
ShangHua-Gao/G2L-search · ShangHua-Gao/RFNext |
2022 |
| 11 |
ConvNext-Tiny (P4, 224x224, SUP) |
48.7 | 48.8 |
A ConvNet for the 2020s
|
keras-team/keras · rwightman/pytorch-image-models · pytorch/vision
· +51 |
2022 |
| 12 |
SERE (ViT-B/16, 100ep, 224x224, SSL) |
48.6 | 48.2 |
SERE: Exploring Feature Self-relation for Self-supervised Transformer
|
MCG-NKU/SERE |
2022 |
| 13 |
TEC (ViT-B/16, 224x224, SSL, mmseg) |
46.1 | 46.0 |
Towards Sustainable Self-supervised Learning
|
sail-sg/tec |
2022 |
| 14 |
TEC (ViT-B/16, 224x224, SSL) |
42.9 | – |
Towards Sustainable Self-supervised Learning
|
sail-sg/tec |
2022 |
| 15 |
SERE (ViT-S/16, 100ep, 224x224, SSL, mmseg) |
41.0 | 40.5 |
SERE: Exploring Feature Self-relation for Self-supervised Transformer
|
MCG-NKU/SERE |
2022 |
| 15 |
SERE (ViT-S/16, 100ep, 224x224, SSL) |
41.0 | 40.2 |
SERE: Exploring Feature Self-relation for Self-supervised Transformer
|
MCG-NKU/SERE |
2022 |
| 17 |
MAE (ViT-B/16, 224x224, SSL, mmseg) |
40.0 | 40.3 |
Masked Autoencoders Are Scalable Vision Learners
|
facebookresearch/mae · lightly-ai/lightly · open-mmlab/mmselfsup
· +55 |
2021 |
| 18 |
MAE (ViT-B/16, 224x224, SSL) |
38.3 | 37.0 |
Masked Autoencoders Are Scalable Vision Learners
|
facebookresearch/mae · lightly-ai/lightly · open-mmlab/mmselfsup
· +55 |
2021 |
| 19 |
PASS (ResNet-50 D16, 224x224, LUSS) |
21.6 | 20.8 |
Large-scale Unsupervised Semantic Segmentation
|
LUSSeg/ImageNet-S · LUSSeg/PASS · LUSSeg/ImageNetSegModel |
2021 |
| 20 |
PASS (ResNet-50 D32, 224x224, LUSS) |
21.0 | 20.3 |
Large-scale Unsupervised Semantic Segmentation
|
LUSSeg/ImageNet-S · LUSSeg/PASS · LUSSeg/ImageNetSegModel |
2021 |