Object Detection on MS-COCO (minival)
63.8APRevCol-H (Objects365+DINO)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RevCol-H (Objects365+DINO)Backbone=RevCol-H, Pre-training=Objects365, Detection Framework=DINO, Params=2.18G, FLOPs=4012G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 63.8 | 81.8 | 70.2 | |
| RevCol-H (HTC++)Backbone=RevCol-H, Pre-training=Extra data, Detection Framework=HTC++, Params=2.41G, FLOPs=4417G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 61.1 | 78.8 | 67 | |
| RevCol-LBackbone=RevCol-L, Pre-training=ImageNet-22K, Params=330M, FLOPs=1453G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 55.9 | 74.1 | 60.7 | |
| RevCol-BBackbone=RevCol-B, Pre-training=ImageNet-22K, Params=196M, FLOPs=988G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 55 | 73.5 | 59.7 | |
| ConvNeXt-LBackbone=ConvNeXt-L, Pre-training=ImageNet-22K, Params=255M, FLOPs=1354G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 54.8 | 73.8 | 59.8 | |
| ConvNeXt-BBackbone=ConvNeXt-B, Pre-training=ImageNet-22K, Params=146M, FLOPs=964G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 54 | 73.1 | 58.8 | |
| Swin-LBackbone=Swin-L, Pre-training=ImageNet-22K, Params=253M, FLOPs=1382G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 53.9 | 72.4 | 58.8 | |
| RepLKNet-LBackbone=RepLKNet-L, Pre-training=ImageNet-22K, Params=229M, FLOPs=1321G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 53.9 | — | — | |
| RevCol-BBackbone=RevCol-B, Pre-training=ImageNet-1K, Params=196M, FLOPs=988G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 53 | 71.4 | 57.3 | |
| Swin-BBackbone=Swin-B, Pre-training=ImageNet-22K, Params=145M, FLOPs=982G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 53 | 71.8 | 57.5 | |
| RepLKNet-BBackbone=RepLKNet-B, Pre-training=ImageNet-22K, Params=137M, FLOPs=965G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 53 | — | — | |
| ConvNeXt-BBackbone=ConvNeXt-B, Pre-training=ImageNet-1K, Params=146M, FLOPs=964G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 52.7 | 71.3 | 57.2 | |
| RevCol-SBackbone=RevCol-S, Pre-training=ImageNet-1K, Params=118M, FLOPs=833G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 52.6 | 71.1 | 56.8 | |
| RepLKNet-BBackbone=RepLKNet-B, Pre-training=ImageNet-1K, Params=137M, FLOPs=965G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 52.2 | — | — | |
| ConvNeXt-SBackbone=ConvNeXt-S, Pre-training=ImageNet-1K, Params=108M, FLOPs=827G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 51.9 | 70.8 | 56.5 | |
| Swin-BBackbone=Swin-B, Pre-training=ImageNet-1K, Params=145M, FLOPs=982G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 51.9 | 70.9 | 56.5 | |
| Swin-SBackbone=Swin-S, Pre-training=ImageNet-1K, Params=107M, FLOPs=838G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 51.8 | 70.4 | 56.3 | |
| RevCol-TBackbone=RevCol-T, Pre-training=ImageNet-1K, Params=88M, FLOPs=741G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 50.6 | 68.9 | 54.9 | |
| Swin-TBackbone=Swin-T, Pre-training=ImageNet-1K, Params=86M, FLOPs=745G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 50.5 | 69.3 | 54.9 | |
| ConvNeXt-TBackbone=ConvNeXt-T, Pre-training=ImageNet-1K, Params=86M, FLOPs=741G, Input Resolution=1280x800, Testing Scale=Single-scale2022.12 | 50.4 | 69.1 | 54.8 | |
| EgoViTWT-allTraining Data=Full Walking Tours dataset2026.03 | 29.6 | — | — | |
| EgoViTZurichTraining Data=WT-Zurich (65-minute)2026.03 | 26.7 | — | — | |
| AttMaskTraining Data=WT-Zurich (65-minute)2026.03 | 25.9 | — | — | |
| MAETraining Data=WT-Zurich (65-minute)2026.03 | 24.6 | — | — | |
| DORATraining Data=WT-Zurich (65-minute)2026.03 | 22.6 | — | — | |
| SimCLRTraining Data=WT-Zurich (65-minute)2026.03 | 22.2 | — | — | |
| iBOTTraining Data=WT-Zurich (65-minute)2026.03 | 22.1 | — | — | |
| DINOTraining Data=WT-Zurich (65-minute)2026.03 | 22 | — | — | |
| MoCo-v3Training Data=WT-Zurich (65-minute)2026.03 | 19 | — | — |