Object Detection on COCO (AP^b, NID)
58AP^bTC-JEPA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TC-JEPABackbone=ViT-L/14, Pre-training Data=CC27M2026.05 | 58 | — | |
| TC-JEPABackbone=ViT-B/16, Pre-training Data=CC27M2026.05 | 55.6 | — | |
| TC-JEPAViT=H/142026.05 | 55.2 | — | |
| TC-JEPABackbone=ViT-B/16, Pre-training Data=YFCC15M2026.05 | 54.5 | — | |
| I-JEPAViT=H/142026.05 | 53.7 | — | |
| StoPViT=H/142026.05 | 53.5 | — | |
| data2vecViT=L/162026.05 | 52.9 | — | |
| SPARCBackbone=ViT-B/16, Pre-training Data=CC27M2026.05 | 52.6 | — | |
| TC-JEPAViT=B/162026.05 | 52.6 | — | |
| SPARCBackbone=ViT-B/16, Pre-training Data=YFCC15M2026.05 | 52 | — | |
| Full-precisionNetwork=Swin-S + Cascade Mask R-CNN, #Bits=32/32, Size=427.82026.05 | 51.9 | — | |
| Full-precisionNetwork=Swin-B + Cascade Mask R-CNN, #Bits=32/32, Size=579.92026.05 | 51.9 | — | |
| RepQ-ViT + NBCNetwork=Swin-S + Cascade Mask R-CNN, #Bits=6/6, Size=91.32026.05 | 51.7 | — | |
| RepQ-ViT + NBCNetwork=Swin-B + Cascade Mask R-CNN, #Bits=6/6, Size=126.12026.05 | 51.7 | — | |
| RepQ-ViTNetwork=Swin-B + Cascade Mask R-CNN, #Bits=6/6, Size=112.12026.05 | 51.5 | — | |
| RepQ-ViTNetwork=Swin-S + Cascade Mask R-CNN, #Bits=6/6, Size=83.42026.05 | 51.4 | — | |
| iBOTViT=B/162026.05 | 51.2 | — | |
| DreamLIPBackbone=ViT-B/16, Pre-training Data=CC30M2026.05 | 50.7 | — | |
| MAEViT=H/142026.05 | 50.1 | — | |
| DINOViT=B/162026.05 | 50.1 | — | |
| RepQ-ViT + NBCNetwork=Swin-B + Cascade Mask R-CNN, #Bits=4/4, Size=90.12026.05 | 49.6 | — | |
| RepQ-ViT + NBCNetwork=Swin-S + Cascade Mask R-CNN, #Bits=4/4, Size=64.82026.05 | 49.4 | — | |
| RepQ-ViTNetwork=Swin-B + Cascade Mask R-CNN, #Bits=4/4, Size=76.12026.05 | 49.3 | — | |
| RepQ-ViTNetwork=Swin-S + Cascade Mask R-CNN, #Bits=4/4, Size=56.92026.05 | 49.1 | — | |
| EvoNet-C3Params=42M, MACs=228G, Lat.=66.9ms, Thrp.=18, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 48.5 | 1.15 | |
| Full-precisionNetwork=Swin-S + Mask R-CNN, #Bits=32/32, Size=276.52026.05 | 48.5 | — | |
| RepQ-ViT + NBCNetwork=Swin-S + Mask R-CNN, #Bits=6/6, Size=61.22026.05 | 47.8 | — | |
| Spatial-Mamba-TParams=46M, MACs=261G, Lat.=51.0ms, Thrp.=20, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 47.6 | 1.03 | |
| RepQ-ViTNetwork=Swin-S + Mask R-CNN, #Bits=6/6, Size=53.32026.05 | 47.6 | — | |
| DreamLIPBackbone=ViT-B/16, Pre-training Data=YFCC15M2026.05 | 47.2 | — | |
| EvoNet-C2Params=36M, MACs=202G, Lat.=55.4ms, Thrp.=23, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 47.1 | 1.29 | |
| LocalVMamba-TParams=45M, MACs=291G, Lat.=108.8ms, Thrp.=8, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 46.7 | 1.04 | |
| VMamba-TParams=42M, MACs=262G, Lat.=74.3ms, Thrp.=12, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 46.5 | 1.11 | |
| MPViT-SParams=43M, MACs=268G, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 46.4 | 1.08 | |
| ViTAS-Twins-SParams=44M, MACs=168G, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 45.9 | 1.04 | |
| EvoNet-C1Params=33M, MACs=190G, Lat.=50.2ms, Thrp.=26, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 45.4 | 1.39 | |
| MaskCLIPBackbone=ViT-B/16, Pre-training Data=YFCC15M2026.05 | 45.4 | — | |
| PVTv2-B2Params=45M, Lat.=86.7ms, Thrp.=7, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 45.3 | 1.01 | |
| ViT-Adapter-SParams=48M, MACs=403G, Lat.=138.7ms, Thrp.=7, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 44.7 | 0.93 | |
| GroupViTBackbone=ViT-S/16, Pre-training Data=CC27M2026.05 | 44.3 | — | |
| ConvNeXt-TParams=48M, MACs=262G, Lat.=43.4ms, Thrp.=26, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 44.2 | 0.92 | |
| CLIPBackbone=ViT-B/16, Pre-training Data=YFCC15M2026.05 | 43.6 | — | |
| BLIPBackbone=ViT-B/16, Pre-training Data=Merged14M2026.05 | 43.1 | — | |
| RepQ-ViT + NBCNetwork=Swin-S + Mask R-CNN, #Bits=4/4, Size=44.02026.05 | 43 | — | |
| RepQ-ViTNetwork=Swin-S + Mask R-CNN, #Bits=4/4, Size=36.12026.05 | 42.8 | — | |
| Swin-TParams=48M, MACs=267G, Lat.=45.1ms, Thrp.=24, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 42.7 | 0.89 | |
| PVT-S+AttnZeroMACs=248G, Lat.=51.3ms, Thrp.=23, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 41 | — | |
| UniNet-B1Params=28M, MACs=150G, Lat.=45.7ms, Thrp.=24, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 40.5 | 1.45 | |
| ResNet-50Params=44M, MACs=260G, Lat.=32.8ms, Thrp.=34, Test Resolution=800 × 1280, Framework=Mask R-CNN [40]2026.03 | 38.2 | 0.87 |