Panoptic Part Segmentation on PASCAL Panoptic Parts (PPP) (val)
60.4PartPQ (All)TAPPS
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TAPPSBackbone=Swin-B, Pre-training=ImageNet + COCO2024.06 | 60.4 | — | — | — | — | — | — | — | — | — | 72.2 | 56.3 | 78.1 | 63 | |
| TAPPSBackbone=ResNet-50, Pre-training=ImageNet + COCO2024.06 | 54.7 | — | — | — | — | — | — | — | — | — | 67.2 | 50.4 | 75.1 | 57.7 | |
| Panoptic-PartFormerBackbone=Swin-B, Pre-training=ImageNet + COCO2024.06 | 54.1 | — | — | — | — | — | — | — | — | — | 64.3 | 50.6 | 70.8 | 58.1 | |
| Panoptic-PartFormer++Backbone=Swin-B, Pre-training=ImageNet + COCO2024.06 | 51.3 | — | — | — | — | — | — | — | — | — | 48.9 | 52.1 | 53 | 59.8 | |
| PPF++Backbone=Swin, Scale=Single-Scale2023.11 | 49.3 | — | — | — | — | — | — | — | — | 52.7 | — | — | — | — | |
| PPF++Backbone=ConvNext, Scale=Single-Scale2023.11 | 48.6 | — | — | — | — | — | — | — | — | 54.2 | — | — | — | — | |
| Panoptic-PartFormer++Backbone=ResNet-50, Pre-training=ImageNet + COCO2024.06 | 45.1 | — | — | — | — | — | — | — | — | — | 52.6 | 42.6 | 60.4 | 51.6 | |
| TAPPSBackbone=ResNet-50, Pre-training=ImageNet2024.06 | 44.6 | — | — | — | — | — | — | — | — | — | 59.6 | 39.4 | 74.3 | 47.1 | |
| Panoptic-PartFormerBackbone=ResNet-50, Pre-training=ImageNet + COCO2024.06 | 43.2 | — | — | — | — | — | — | — | — | — | 56.1 | 38.8 | 66.8 | 47.6 | |
| PPF++Backbone=ResNet101, Scale=Single-Scale2023.11 | 42.4 | — | — | — | — | — | — | — | — | 46 | — | — | — | — | |
| PPF++Backbone=ResNet50, Scale=Single-Scale2023.11 | 42.2 | — | — | — | — | — | — | — | — | 45.2 | — | — | — | — | |
| JPPFBackbone=EfficientNet-L2, Scale=Single-Scale2023.11 | 40.5 | — | — | — | — | — | 58.5 | 34.4 | — | 53.7 | — | — | — | — | |
| PPFBackbone=ResNet101, Scale=Single-Scale2023.11 | 39.3 | — | — | — | — | — | — | — | — | 41.3 | — | — | — | — | |
| Baseline-2Scale=Single-Scale, Model size [M]=1112022.12 | 38.3 | — | — | 55.1 | 44.8 | 58.6 | 51.6 | 33.8 | — | — | — | — | — | — | |
| Baseline-2Backbone=ResNeSt269, Scale=Single-Scale2023.11 | 38.3 | — | — | — | — | — | 51.6 | 33.8 | — | — | — | — | — | — | |
| PPFBackbone=ResNet50, Scale=Single-Scale2023.11 | 37.8 | — | — | — | — | — | — | — | — | 40.2 | — | — | — | — | |
| JPPFScale=Single-Scale, Run time [ms]=146, Model size [M]=44.192022.12 | 32.3 | — | — | 46 | 39.1 | 54.4 | 48.3 | 26.9 | 92.1 | — | — | — | — | — | |
| JPPFBackbone=EfficientNet-B5, Scale=Single-Scale2023.11 | 32.3 | — | — | — | — | — | 48.3 | 26.9 | — | 45.6 | — | — | — | — | |
| JPPFBackbone=EfficientNet-B5, Pre-training=ImageNet2024.06 | 32.2 | — | — | — | — | — | — | — | — | — | 48.3 | 26.9 | — | — | |
| Baseline-1Scale=Single-Scale, Model size [M]=68 (encoders only)2022.12 | 31.4 | — | — | 47.1 | 38.5 | 53.9 | 47.2 | 26 | — | — | — | — | — | — | |
| Baseline-1Backbone=ResNet50, Scale=Single-Scale2023.11 | 31.4 | — | — | — | — | — | 47.2 | 26 | — | — | — | — | — | — | |
| DeepLabv3+ & Mask R-CNN | DeepLabv3+Panoptic seg. method=DeepLabv3+ & Mask R-CNN [18], Part seg. method=DeepLabv3+ [3], Backbone=ResNet50, Inference scale=single scale2022.04 | — | 0.35 | 0.314 | — | — | — | — | — | — | — | — | — | — | — | |
| DLv3-ResNeSt269 & DetectoRS | BSANetPanoptic seg. method=DLv3-ResNeSt269 [71] & DetectoRS [49], Part seg. method=BSANet [75], Inference scale=single scale2022.04 | — | 0.42 | 0.383 | — | — | — | — | — | — | — | — | — | — | — | |
| HIPIESegmentation Mode=Cls-aware, Open Vocabulary=true, Unified Framework=true2023.07 | — | — | — | — | — | 5.2 | — | — | — | — | — | — | — | — | |
| Panoptic-PartFormerBackbone=ResNet50, Inference scale=single scale2022.04 | — | 0.476 | 0.378 | — | — | — | — | — | — | — | — | — | — | — | |
| Panoptic-PartFormerBackbone=ResNet101, Inference scale=single scale2022.04 | — | 0.492 | 0.393 | — | — | — | — | — | — | — | — | — | — | — |