search

LiUzHiAn / hf2vad

MIT License
124 stars 24 forks source link

The pretrained model output AUC=0.5? #25

Open DaffTangEclipse opened 2 years ago

DaffTangEclipse commented 2 years ago

I use the same original dataset ped2, produce the flow, and the pretrained model, but both eval.py and ml_memAE_sc_eval.py output AUC=0.5, I'm confused. I debug the code, saw that a variable frame_bbox_scores in eval.py -> evaluate() only has data for the first 20.I don't know why.... frame_bbox_score

LiUzHiAn commented 2 years ago

I guess there might be something missing from the data preporcessing. The frame_bbox_scores contain anomaly scores for each object within each frame. From the output you gave, it seems that you only extract spatial-temporal cubes for the first 20 frames, while the samples for the rest frames are empty.

Could you please check if the variable scores(here) is empty for every batch? For data preprocessing correctness validation, I plan to share the processed file I used (for ped2), this would be completed in a few days.