search

yifanlu0227 / HEAL

[ICLR2024] HEAL: An Extensible Framework for Open Heterogeneous Collaborative Perception ➡️ All You Need for Multi-Modality Collaborative Perception!
Other
138 stars 9 forks source link

The no collaboration experiment based on dair_v2x dataset #23

Open KKHAOO opened 1 month ago

KKHAOO commented 1 month ago

I train with this yaml https://github.com/yifanlu0227/HEAL/blob/main/opencood/hypes_yaml/dairv2x/Single/DAIR_single_m1.yaml and infer with --fusion_method no However, my results are significantly different from yours. I would like to ask if what I am doing is correct. How can I obtain the same results as yours? this is my result: image and this is yours: 1721012430827

yifanlu0227 commented 1 month ago

Hi! I also run the no-collaboration experiment again, and the result turns out to be similar as yours.

And In my checkpoints (where I also upload the metric and visualization results), I just found I set a smaller detection range --range 102.4,51.2 for DAIR-V2X's LiDAR baseline experiments. But the AP performance would not change too much, exactly higher than the result in the table.

I don't have my previous log for this experiment. Your current result should be correct.

yifanlu0227 commented 1 month ago

And to be honest, if you infer with --fusion_method late, you can get a even higher AP score in DAIR-V2X dataset though there are many false positives.