megvii-research / Iter-E2EDET

Official implementation of the paper "Progressive End-to-End Object Detection in Crowded Scenes"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to obtain the same accuracy as the paper?

ma3252788 opened this issue · comments

Hello, may I ask why, according to your instructions, when I directly run this command python3 train_net.py --num-gpus 1 --config-file configs/50e.6h.500pro.ignore.yaml --eval-only MODEL.WEIGHTS ../../e2edet_final.pth, there is only AP value and no MR. in addition, the AP value is also very low, only 82%

Thank you very much for any reply

image

commented

emm, please run demo.py in each project for evaluation.

Thank you for your reply, but it seems that there is no demo.py file in the project

image

commented

You can download the provided model for fast evaluation, whose performance is reported in our paper. Also, we use Pascal-VOC style evaluation code for $\text{AP}$ and $\text{MR}^{-2}$ evaluation, which is a bit different from that in the COCO. Sorry, the demo.py is renamed as eval_result.py when we shared the code.

image