Inference Performance on UniAD
1.27Latency (ms)M100
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| M100Network Model=TrackFormer, Active Clusters=82026.04 | 1.27 | — | 6.3 | |
| M100Network Model=MapFormer, Active Clusters=82026.04 | 1.46 | — | 4.2 | |
| M100Network Model=FPN, Active Clusters=82026.04 | 4.23 | — | 1.2 | |
| M100Network Model=TempFusion, Active Clusters=82026.04 | 4.47 | — | 3.8 | |
| Thor-UNetwork Model=FPN2026.04 | 5.1 | — | — | |
| Thor-UNetwork Model=MapFormer2026.04 | 6.14 | — | — | |
| M100Network Model=BEVFormer, Active Clusters=82026.04 | 7.92 | — | 4.1 | |
| Thor-UNetwork Model=TrackFormer2026.04 | 7.95 | — | — | |
| M100Network Model=RegNet, Active Clusters=82026.04 | 13.1 | — | 4.4 | |
| Thor-UNetwork Model=TempFusion2026.04 | 17 | — | — | |
| Thor-UNetwork Model=BEVFormer2026.04 | 32.83 | — | — | |
| Thor-UNetwork Model=RegNet2026.04 | 57.4 | — | — | |
| M100Metric Category=Total Perception Frame rate, Active Clusters=82026.04 | — | 30 | 3.8 | |
| Thor-UMetric Category=Total Perception Frame rate2026.04 | — | 7.9 | — |