Trajectory Prediction on Proprietary Driving Dataset held-out (test)
0.761minADE (6)Flex
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FlexLLM=Qwen2-0.5B, Stage=Stage 1 + Stage 2, #Camera×#Time=2×9, #Scene tokens=9002025.12 | 0.761 | 3,318 | 41.08 | |
| FlexLLM=Qwen2-0.5B, Stage=Stage 1, #Camera×#Time=2×9, #Scene tokens=9002025.12 | 0.794 | 2,260 | 41.08 | |
| In-house VLALLM=Qwen2-0.5B, Stage=Stage 1 + Stage 2, #Camera×#Time=2×9, #Scene tokens=28802025.12 | 0.798 | 5,750 | 18.6 | |
| In-house VLALLM=Qwen2-0.5B, Stage=Stage 1, #Camera×#Time=2×9, #Scene tokens=28802025.12 | 0.818 | 4,134 | 18.6 | |
| Triplane (4-6-6)LLM=Llama2-1B, Stage=Stage 1, #Camera×#Time=4×6, #Scene tokens=24962025.12 | 0.974 | 7,960 | 64.53 | |
| Triplane (8-8-8)LLM=Llama2-1B, Stage=Stage 1, #Camera×#Time=4×6, #Scene tokens=10802025.12 | 1.046 | 4,984 | 69.62 |