| Rank | Model |
PQ | mIoU | PQst | PQth |
Paper | Code | Year |
| 1 |
OneFormer (DiNAT-L, single-scale) |
46.7 | 61.7 | 54.9 | 40.5 |
OneFormer: One Transformer to Rule Universal Image Segmentation
|
huggingface/transformers · SHI-Labs/OneFormer · yangyucheng000/University
· +1 |
2022 |
| 2 |
OneFormer (ConvNeXt-L, single-scale) |
46.4 | 61.6 | 54.0 | 40.6 |
OneFormer: One Transformer to Rule Universal Image Segmentation
|
huggingface/transformers · SHI-Labs/OneFormer · yangyucheng000/University
· +1 |
2022 |
| 3 |
Panoptic FCN* (Swin-L, single-scale) |
45.7 | – | 52.1 | 40.8 |
Fully Convolutional Networks for Panoptic Segmentation
|
dvlab-research/panopticfcn · yanwei-li/PanopticFCN · Jia-Research-Lab/PanopticFCN
· +3 |
2020 |
| 4 |
Panoptic-DeepLab (SWideRNet-(1, 1, 4.5), multi-scale) |
44.8 | 60.0 | 51.9 | 39.3 |
Scaling Wide Residual Networks for Panoptic Segmentation
|
|
2020 |
| 5 |
Mask2Former + Intra-Batch Supervision (ResNet-50) |
42.2 | – | 52.0 | 34.9 |
Intra-Batch Supervision for Panoptic Segmentation on High-Resolution Images
|
DdeGeus/Mask2Former-IBS · DdeGeus/PanopticFCN-IBS |
2023 |
| 6 |
Axial-DeepLab-L (multi-scale) |
41.1 | 58.4 | 51.3 | 33.4 |
Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation
|
The-AI-Summer/self_attention · google-research/deeplab2 · csrhddlam/axial-deeplab
· +2 |
2020 |
| 7 |
EfficientPS |
40.6 | – | – | – |
EfficientPS: Efficient Panoptic Segmentation
|
DeepSceneSeg/EfficientPS · vincrichard/EfficientPS |
2020 |
| 8 |
Panoptic-DeepLab (X71) |
40.5 | – | – | – |
Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation
|
tensorflow/models · PaddlePaddle/PaddleSeg · google-research/deeplab2
· +6 |
2019 |
| 9 |
AdaptIS (ResNeXt-101) |
40.3 | 56.8 | – | – |
AdaptIS: Adaptive Instance Selection Network
|
|
2019 |
| 10 |
Panoptic FCN* (ResNet-FPN) |
36.9 | – | – | 32.9 |
Fully Convolutional Networks for Panoptic Segmentation
|
dvlab-research/panopticfcn · yanwei-li/PanopticFCN · Jia-Research-Lab/PanopticFCN
· +3 |
2020 |
| 11 |
JSIS-Net (ResNet-50) |
17.6 | – | – | – |
Panoptic Segmentation with a Joint Semantic and Instance Segmentation Network
|
|
2018 |
| 11 |
HRNet-OCR (Hierarchical Multi-Scale Attention) |
17.6 | – | – | – |
Hierarchical Multi-Scale Attention for Semantic Segmentation
|
PaddlePaddle/PaddleSeg · NVIDIA/semantic-segmentation · Song-Jingyu/runnable-PointPainting
· +5 |
2020 |
| 13 |
Panoptic FCN* (ResNet-50-FPN) |
– | – | 42.3 | – |
Fully Convolutional Networks for Panoptic Segmentation
|
dvlab-research/panopticfcn · yanwei-li/PanopticFCN · Jia-Research-Lab/PanopticFCN
· +3 |
2020 |
| 14 |
OneFormer (DiNAT-L, single-scale) |
46.7 | 61.7 | 54.9 | 40.5 |
OneFormer: One Transformer to Rule Universal Image Segmentation
|
huggingface/transformers · SHI-Labs/OneFormer · yangyucheng000/University
· +1 |
2022 |
| 15 |
OneFormer (ConvNeXt-L, single-scale) |
46.4 | 61.6 | 54.0 | 40.6 |
OneFormer: One Transformer to Rule Universal Image Segmentation
|
huggingface/transformers · SHI-Labs/OneFormer · yangyucheng000/University
· +1 |
2022 |
| 16 |
Panoptic FCN* (Swin-L, single-scale) |
45.7 | – | 52.1 | 40.8 |
Fully Convolutional Networks for Panoptic Segmentation
|
dvlab-research/panopticfcn · yanwei-li/PanopticFCN · Jia-Research-Lab/PanopticFCN
· +3 |
2020 |
| 17 |
Panoptic-DeepLab (SWideRNet-(1, 1, 4.5), multi-scale) |
44.8 | 60.0 | 51.9 | 39.3 |
Scaling Wide Residual Networks for Panoptic Segmentation
|
|
2020 |
| 18 |
Mask2Former + Intra-Batch Supervision (ResNet-50) |
42.2 | – | 52.0 | 34.9 |
Intra-Batch Supervision for Panoptic Segmentation on High-Resolution Images
|
DdeGeus/Mask2Former-IBS · DdeGeus/PanopticFCN-IBS |
2023 |
| 19 |
Axial-DeepLab-L (multi-scale) |
41.1 | 58.4 | 51.3 | 33.4 |
Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation
|
The-AI-Summer/self_attention · google-research/deeplab2 · csrhddlam/axial-deeplab
· +2 |
2020 |
| 20 |
EfficientPS |
40.6 | – | – | – |
EfficientPS: Efficient Panoptic Segmentation
|
DeepSceneSeg/EfficientPS · vincrichard/EfficientPS |
2020 |