AP evaluation on the nuScenes dataset #32
Description
Hi, thank you so much for sharing such a comprehensive and well-structured codebase — it's been incredibly helpful for my research!
I wanted to ask about the AP evaluation on the nuScenes dataset. I'm currently getting all-zero AP results when running the evaluation script, and I was wondering if you could share the expected AP performance of DEVIANT on nuScenes, just to make sure my setup is correct.
Also, if possible, could you provide a sample submission.json file (even for a single example) so I can double-check the format and structure?
From the training logs, the losses seem to be converging normally. I've already updated the paths in rpn_util.py, and the submission.json file is successfully generated. When I run the evaluation script with --verbose 1, I can see that many predictions remain after filtering, but the final mAP is always zero.
If there’s anything else you think I should check, I’d really appreciate your guidance. Thanks again for your great work!