search

DerrickXuNu / OpenCOOD

[ICRA 2022] An opensource framework for cooperative detection. Official implementation for OPV2V.
https://mobility-lab.seas.ucla.edu/opv2v/
Other
644 stars 99 forks source link

A sharp decline in performance from OPV2V to V2Xset? #114

Open lubin2022 opened 10 months ago

lubin2022 commented 10 months ago

I want to train the pointpillar_v2vnet model on both OpV2V and V2XSet dataset, the yaml file are the same except for the 'root_dir' and 'validate_dir'. One is 'opv2v/train', 'opv2v/validate' and the other one is 'v2xset/train', 'v2xset/validate'. For the OPV2V dataset, AP@0.5=0.914, AP@0.7=0.798, but for the V2XSet dataset, AP@0.5=0.672, AP@0.7=0.315 Why does the performance differ so much between two datasets? Is it normal?

DerrickXuNu commented 9 months ago

V2XSet is usually more challenging than opv2v, but from my own experiments, the difference is not that large as yours. I think you didn't train the models well on V2XSet