bug report for bestcheckpoint.pth.tar

[kaenkogashi]

Dear Sir,

Thanks for your excellent work! I followed the instructions and trained the COCO model from scratch. After I got bestcheckpoint.pth.tar, I tried to store the results in v-coco format, but I got "Failed to load checkPoint" error message.

~/VSGNet/scripts$ CUDA_VISIBLE_DEVICES=0,1,2,3 python2 main.py -fw COCO_result -ba 8 -r t -i t Doing a test run to detect bad detections

In training set object detector failed to detect any person in the following images: [6357, 17487, 130851, 153892, 176148, 326601, 364400, 365013, 369213, 429514, 483039, 560726] In validation set object detector failed to detect any person in the following images: [130099, 196981, 352877, 387895, 426849, 526087, 567439, 568117] In testing set object detector failed to detect any person in the following images: [16875, 30828, 56701, 66706, 75768, 165157, 228418, 230501, 255483, 267725, 293855, 344045, 413805, 419143, 451038, 479280, 561411] ../COCO_result/bestcheckpoint.pth.tar 0 Failed to load checkPoint Only doing testing for storing result from a model Epoch 0/0

I noticed bestcheckpoint.pth.tar's data size(303633714 ) is a little different from other tar such as 20checkpoint.pth.tar(302550252 ). Is this is a bug or not? Thank you very much! I only got "Failed to load checkPoint error " by bestcheckpoint.pth.tar. Other tars are worked.

-rw-rw-r-- 1 776297312 7月 30 21:44 test20.pickle -rw-rw-r-- 1 302550252 7月 30 21:44 20checkpoint.pth.tar -rw-rw-r-- 1 774514896 7月 31 01:40 test40.pickle -rw-rw-r-- 1 302550252 7月 31 01:40 40checkpoint.pth.tar -rw-rw-r-- 1 771836105 7月 31 05:40 test60.pickle -rw-rw-r-- 1 302550252 7月 31 05:40 60checkpoint.pth.tar -rw-rw-r-- 1 303633714 7月 31 07:50 bestcheckpoint.pth.tar -rw-rw-r-- 1 768390928 7月 31 09:44 test80.pickle -rw-rw-r-- 1 302550252 7月 31 09:44 80checkpoint.pth.tar -rw-rw-r-- 1 6 7月 31 09:44 result.pickle

[kaenkogashi]

Already solved by myself. thank you very much!