ifzhang / FairMOT

[IJCV-2021] FairMOT: On the Fairness of Detection and Re-Identification in Multi-Object Tracking
MIT License
4.03k stars 934 forks source link

Has anyone ever run fairmot's dla network in CMU panoptic? #533

Open iwsuccess opened 1 year ago

iwsuccess commented 1 year ago

I modified the heatmap head to predict the heatmaps of 15 joint points. I use MSE loss. Loss can converge to about 0.0002 on the training set, and about 0.001 on the verification set. but after I visualized the prediction results of heatmaps, I found that the visual result is not good. I don't know why? Please consult everyone! heatmap GT: NO_386_ViewNo_2_hm_gt heatmap pred: NO_386_ViewNo_2_hm_pred