How to run the evaluation for Dolphins and DriveLM?

[JasonBourne1998]

Congratulations on the ICCV'25 acceptance. May I know how you implement the code for Dolphins and DriveLM? Do you treat them as E2E model or just for downstream tasks？

[Daniel-xsy]

Hi @JasonBourne1998, Thanks for your interest. We only evaluate the language output on our dataset. Thus, no E2E evaluation is included in our work (e.g., waypoint prediction).

For DriveLM, please refer to this issue: #7.
For Dolphins, we use their uploaded weight on HuggingFace.

We also upload our unorganized evaluation script for other models in Google Drive.

[pzhwuhu]

I'm a begginner with this. For Dolphins, I can't find the "uploaded weight" in the provided url, could you possibly verify and explicitly indicate where it is ?

[pzhwuhu]

Oh, I found it in the source code, this is the weight url, thanks ！