Image Tokenization on nuScenes
27.91PSNRFlowMo-Hi
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FlowMo-HiInput=256 × 2562026.03 | 27.91 | 79.3 | 259.7 | 8,436.6 | |
| DriveTokInput=6 × 256 × 7042026.03 | 27.89 | 74.7 | 87.96 | 4,423.73 | |
| DriveTokInput=6 × 256 × 2562026.03 | 27.13 | 72.3 | 21.86 | 3,957.95 | |
| ViT-VQGANInput=256 × 2562026.03 | 26.96 | 76.3 | 59.23 | 1,354.4 | |
| VQGAN†Input=256 × 256, Training Data=OpenImages [25]2026.03 | 26.77 | 75.6 | 63.34 | 1,440.7 | |
| BEV-VAEInput=6 × 256 × 2562026.03 | 26.32 | 74.6 | — | — | |
| VQGAN‡Input=256 × 256, Training Data=Imagenet [7]2026.03 | 24.41 | 67.4 | 63.31 | 1,440.7 | |
| FlowMo-LoInput=256 × 2562026.03 | 23.11 | 69.6 | 248.04 | 8,434.8 |