Instance Segmentation on LVIS v1 (val)
35.5AP (m, r)FRACAL
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FRACALBackbone=Swin-B, Schedule=1x, Architecture=Mask R-CNN2024.10 | 35.5 | — | — | — | — | — | 38.5 | 39.4 | 38.7 | — | — | — | |
| DenseBackbone=ViT-H, PR (%)=02026.02 | 34.61 | 44.65 | 32.09 | 57.54 | 65.68 | — | — | 45.43 | 48.19 | 61.15 | 47.38 | 48.53 | |
| POPBackbone=ViT-H, PR (%)=202026.02 | 34.14 | 43.66 | 31.76 | 55.77 | 63.87 | — | — | 44.31 | 47.13 | 60.68 | 46.37 | 47.52 | |
| RandomBackbone=ViT-H, PR (%)=202026.02 | 34.1 | 42.82 | 30.23 | 55.26 | 64.07 | — | — | 43.55 | 45.86 | 60.06 | 45.37 | 46.81 | |
| POPBackbone=ViT-L, PR (%)=202026.02 | 33.87 | 43.34 | 31.83 | 55.43 | 63.01 | — | — | 44.2 | 46.54 | 60.58 | 46.03 | 47.2 | |
| DenseBackbone=ViT-L, PR (%)=02026.02 | 33.84 | 44.01 | 31.48 | 56.77 | 65.46 | — | — | 44.94 | 47.44 | 60.78 | 46.63 | 47.93 | |
| DenseBackbone=ViT-B, PR (%)=02026.02 | 32.96 | 41.41 | 29.47 | 53.49 | 61.39 | — | — | 42.27 | 44.18 | 59.61 | 43.68 | 45.38 | |
| RandomBackbone=ViT-L, PR (%)=202026.02 | 32.75 | 42.26 | 29.7 | 54.79 | 64.15 | — | — | 43.31 | 45.26 | 59.72 | 44.56 | 46.28 | |
| POPBackbone=ViT-B, PR (%)=202026.02 | 32.6 | 41.07 | 29.56 | 52.87 | 60.49 | — | — | 41.99 | 43.76 | 59.28 | 43.51 | 45.02 | |
| RandomBackbone=ViT-B, PR (%)=202026.02 | 31.05 | 39.2 | 27.61 | 50.65 | 59.1 | — | — | 40.5 | 41.34 | 58 | 40.9 | 43.15 | |
| BaselineBackbone=Swin-B, Schedule=1x, Architecture=Mask R-CNN2024.10 | 28.9 | — | — | — | — | — | 36.6 | 37.8 | 38.7 | — | — | — | |
| FRACALBackbone=Swin-S, Schedule=1x, Architecture=Mask R-CNN2024.10 | 27.8 | — | — | — | — | — | 33.6 | 33.9 | 35.9 | — | — | — | |
| FRACALBackbone=Swin-T, Schedule=1x, Architecture=Mask R-CNN2024.10 | 25.7 | — | — | — | — | — | 30.7 | 30.5 | 33.2 | — | — | — | |
| SeesawBackbone=Swin-S, Schedule=1x, Architecture=Mask R-CNN2024.10 | 25.6 | — | — | — | — | — | 32.4 | 32.8 | 34.9 | — | — | — | |
| FRACALArch.=Mask RCNN ResNet1012024.10 | 24.5 | — | — | — | — | — | 29.8 | 29.3 | 32.7 | — | — | — | |
| GOLBackbone=Swin-S, Schedule=1x, Architecture=Mask R-CNN2024.10 | 24.1 | — | — | — | — | — | 31.5 | 32.3 | 33.8 | — | — | — | |
| SeesawBackbone=Swin-T, Schedule=1x, Architecture=Mask R-CNN2024.10 | 24 | — | — | — | — | — | 29.5 | 29.3 | 32.2 | — | — | — | |
| FRACALArch.=Mask RCNN ResNet502024.10 | 23 | — | — | — | — | — | 28.6 | 28 | 31.5 | — | — | — | |
| LogNArch.=Mask RCNN ResNet1012024.10 | 22.9 | — | — | — | — | — | 29 | 28.8 | 31.8 | — | — | — | |
| GOLArch.=Mask RCNN ResNet1012024.10 | 22.8 | — | — | — | — | — | 29 | 29 | 31.7 | — | — | — | |
| CRAT w/ LOCEArch.=Mask RCNN ResNet1012024.10 | 22 | — | — | — | — | — | 28.8 | 28.6 | 32 | — | — | — | |
| ECMArch.=Mask RCNN ResNet1012024.10 | 21.9 | — | — | — | — | — | 28.7 | 27.9 | 32.3 | — | — | — | |
| LogNArch.=Mask RCNN ResNet502024.10 | 21.8 | — | — | — | — | — | 27.5 | 27.1 | 30.4 | — | — | — | |
| BaselineBackbone=Swin-S, Schedule=1x, Architecture=Mask R-CNN2024.10 | 21.7 | — | — | — | — | — | 30.9 | 31 | 34.7 | — | — | — | |
| GOLArch.=Mask RCNN ResNet502024.10 | 21.4 | — | — | — | — | — | 27.7 | 27.7 | 30.4 | — | — | — | |
| CRAT w/ LOCEArch.=Mask RCNN ResNet502024.10 | 21.2 | — | — | — | — | — | 27.5 | 26.8 | 31 | — | — | — | |
| GOLBackbone=Swin-T, Schedule=1x, Architecture=Mask R-CNN2024.10 | 21.1 | — | — | — | — | — | 28.5 | 29.5 | 30.6 | — | — | — | |
| ROGArch.=Mask RCNN ResNet1012024.10 | 21.1 | — | — | — | — | — | 28.8 | 29.1 | 31.8 | — | — | — | |
| NorCalArch.=Mask RCNN ResNet1012024.10 | 20.8 | — | — | — | — | — | 27.3 | 26.5 | 31 | — | — | — | |
| ECMArch.=Mask RCNN ResNet502024.10 | 19.7 | — | — | — | — | — | 27.4 | 27 | 31.1 | — | — | — | |
| NorCalArch.=Mask RCNN ResNet502024.10 | 19.3 | — | — | — | — | — | 25.2 | 24.2 | 29 | — | — | — | |
| BaselineBackbone=Swin-T, Schedule=1x, Architecture=Mask R-CNN2024.10 | 17.9 | — | — | — | — | — | 27.7 | 27.9 | 31.8 | — | — | — | |
| BaselineArch.=Mask RCNN ResNet1012024.10 | 16.8 | — | — | — | — | — | 27 | 26.5 | 32 | — | — | — | |
| BaselineArch.=Mask RCNN ResNet502024.10 | 15.8 | — | — | — | — | — | 25.7 | 25.1 | 30.6 | — | — | — | |
| EfficientViT-SAM-L0Params=35M, MACs=35G, Throughput (A100 image/s)=762, Evaluation Mode=Zero-Shot, Prompt=ViTDet Boxes, Precision=fp16, Engine=TensorRT2022.05 | — | 41.8 | 28.8 | 53.4 | 64.7 | — | — | — | — | — | — | — | |
| EfficientViT-SAM-L1Params=48M, MACs=49G, Throughput (A100 image/s)=638, Evaluation Mode=Zero-Shot, Prompt=ViTDet Boxes, Precision=fp16, Engine=TensorRT2022.05 | — | 42.1 | 29.1 | 54.3 | 65 | — | — | — | — | — | — | — | |
| EfficientViT-SAM-L2Params=61M, MACs=69G, Throughput (A100 image/s)=538, Evaluation Mode=Zero-Shot, Prompt=ViTDet Boxes, Precision=fp16, Engine=TensorRT2022.05 | — | 42.7 | 29.4 | 55.1 | 65.5 | — | — | — | — | — | — | — | |
| EfficientViT-SAM-XL0Params=117M, MACs=185G, Throughput (A100 image/s)=278, Evaluation Mode=Zero-Shot, Prompt=ViTDet Boxes, Precision=fp16, Engine=TensorRT2022.05 | — | 43.9 | 31.2 | 56.2 | 65.9 | — | — | — | — | — | — | — | |
| EfficientViT-SAM-XL1Params=203M, MACs=322G, Throughput (A100 image/s)=182, Evaluation Mode=Zero-Shot, Prompt=ViTDet Boxes, Precision=fp16, Engine=TensorRT2022.05 | — | 44.4 | 31.6 | 57 | 66.4 | — | — | — | — | — | — | — | |
| MAEDataset=IN1k, Arch.=ViTDet-H, Framework=Cascade2023.03 | — | — | — | — | — | 46.6 | — | — | — | — | — | — | |
| Mask R-CNNFramework=Mask R-CNN, ProCo=false, Backbone=ResNet-50, Schedule=1x2024.03 | — | — | — | 21.7 | — | — | — | — | — | — | — | — | |
| Mask R-CNN + ProCoFramework=Mask R-CNN, ProCo=true, Backbone=ResNet-50, Schedule=1x2024.03 | — | — | — | 24.7 | — | — | — | — | — | — | — | — | |
| MAWSDataset=IG-3B, Arch.=ViTDet-H, Framework=Cascade2023.03 | — | — | — | — | — | 45.5 | — | — | — | — | — | — | |
| MAWSDataset=IG-3B, Arch.=ViTDet-2B, Framework=Cascade2023.03 | — | — | — | — | — | 46.1 | — | — | — | — | — | — | |
| SAMSupervision protocol=zero-shot, Prompting source=ViTDet boxes2023.04 | — | 44.7 | 32.5 | 57.6 | 65.5 | — | — | — | — | — | — | — | |
| SAM-VIT-HParams=641M, MACs=2973G, Throughput (A100 image/s)=11, Evaluation Mode=Zero-Shot, Prompt=ViTDet Boxes, Precision=fp16, Engine=TensorRT2022.05 | — | 44.2 | 31.8 | 57.1 | 65.3 | — | — | — | — | — | — | — | |
| SWAGDataset=IG-3.6B, Arch.=ViTDet-H, Framework=Cascade2023.03 | — | — | — | — | — | 42.1 | — | — | — | — | — | — | |
| ViTDet-HSupervision protocol=fully-supervised2023.04 | — | 46.6 | 35 | 58 | 66.3 | — | — | — | — | — | — | — | |
| Winner 2021Dataset=IN21k, Arch.=CBNetv2, Framework=HTC2023.03 | — | — | — | — | — | 49.2 | — | — | — | — | — | — |