search

megvii-research / Iter-E2EDET

Official implementation of the paper "Progressive End-to-End Object Detection in Crowded Scenes"
MIT License
89 stars 8 forks source link

Reproduce Baseline Results #30

Open mdaehl opened 5 months ago

mdaehl commented 5 months ago

Hello, I would like to reproduce your baseline results (specifically Deformable DETR on Crowdhuman). Could you describe your trainig settings such as LR, Augmentation etc. including the usage of box refinement/two stage or point me towards the config you used to achieve the results? I only found a detailed description for the Deformable DETR version used in combination with your idea in the supplementary of your paper.

Also you said in #11 that you used a VOC style evaluation for $MR^{-2}$, but in the code it looks llike you always use the caltech compare method instead. Am I overlooking something or did you use the caltch comparison method for your final results?

Thanks already in advance!

yexiguafuqihao commented 4 months ago

This project could help you.