search

zhenghao977 / FCOS-PyTorch-37.2AP

A pure torch implement of FCOS 37.2AP
162 stars 39 forks source link

0it [00:00, ?it/s] #2

Closed shaojiestar closed 3 years ago

shaojiestar commented 3 years ago

when I give the COCOGenerator the path of coco test2017 and image_info_test-dev2017.json.the coco_eval.py can't run normally。and there are not error info when I add a line to show the len of generator,it's 0 below is the output of the terminal: (BorderDet) PS \FCOS-PyTorch-37.2AP-master> python -u "\FCOS-PyTorch-37.2AP-master\coco_eval.py" loading annotations into memory... Done (t=0.08s) creating index... index created! INFO====>check annos, filtering invalid data...... INFO===>success frozen BN INFO===>success frozen backbone stage1 0 0it [00:00, ?it/s]

zhenghao977 commented 3 years ago

I didn't meet this problem, in my env, the codes run well. I suggest you check your json file.

shaojiestar commented 3 years ago

I have solved this problem.there are no annitions in MS COCO test-dev dataset.So the program can't read any annitions in the json file. In the end, I just revise your code so that I can predict the images in test-dev and generate the json file.Then I upload this test-dev result json file to the MS COCO detection(bbox) evaluation server,I get the map(@0.5.0.95)=0.384

zhenghao977 commented 3 years ago

I see. the ap which is a little lower than paper may because of the difference of the post-processing.