Monocular 3D Object Detection on nuScenes (val)
32mAPSparseViT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SparseViTResolution=288x792, Width=1x, MACs (G)=113.9, Latency (ms)=34.52023.03 | 32 | 72.8 | 27.2 | 53.8 | 79.4 | 25.7 | 40.1 | |
| FCOS3D-SW-smallLr schd=1x2024.01 | 31.42 | — | — | — | — | — | — | |
| FCOS3D-UniRepLKNet-SLr schd=1x2024.01 | 31.38 | — | — | — | — | — | — | |
| Swin-TResolution=256x704, Width=1x, MACs (G)=140.8, Latency (ms)=36.42023.03 | 31.2 | 69.1 | 27.2 | 52.3 | 90.9 | 24.7 | 39.2 | |
| SparseViTResolution=256x704, Width=1x, MACs (G)=78.4, Latency (ms)=23.82023.03 | 31.2 | 70.9 | 27.5 | 58.7 | 83.1 | 27.2 | 38.9 | |
| SparseViTResolution=256x704, Width=1x, MACs (G)=58.6, Latency (ms)=18.72023.03 | 30 | 72 | 27.5 | 59.7 | 81.7 | 26.6 | 38.3 | |
| Swin-T (R224)Resolution=224x616, Width=1x, MACs (G)=78.5, Latency (ms)=232023.03 | 29.9 | 71.8 | 27.4 | 60.9 | 79 | 26 | 38.4 | |
| Swin-T (W0.6x)Resolution=256x704, Width=0.6x, MACs (G)=56, Latency (ms)=22.62023.03 | 29.9 | 69.9 | 27.5 | 59.9 | 81.4 | 25.8 | 38.5 | |
| FCOS3D-ResNet101 w/ DCNLr schd=1x2024.01 | 29.8 | — | — | — | — | — | — | |
| Swin-T (R192)Resolution=192x528, Width=1x, MACs (G)=67.1, Latency (ms)=18.72023.03 | 28.7 | 74.3 | 27.9 | 59.5 | 76.7 | 27.8 | 37.7 | |
| Swin-T (W0.4x)Resolution=256x704, Width=0.4x, MACs (G)=20.4, Latency (ms)=17.62023.03 | 27.6 | 74.2 | 27.9 | 63.4 | 91 | 26.2 | 35.5 |