Request: Release evaluation code for L2 1s/2s/3s and Failure rate (≥10 m) in Table 1

[ShuntaroItakura]

Thank you for the excellent work!

Could you please release the evaluation script used to compute the results in Table 1? Specifically:

• L2 (m) 1s, 2s, 3s, ave
• Failure rate (%) with a 10 m threshold

Validation results in my environment

When I evaluated Qwen2.0-VL-7B on the nuScenes Validation split and applied the 10 m threshold, I observed only one failure out of 150 scenes — specifically, scene 0636.

I would like to understand where the difference between my results and the paper’s Table 1 might be coming from.

Thank you very much in advance.