a test problem about centernet

[ZhenLiuBuaa]

so,the output file results.json is a empty list

[ZhenLiuBuaa]

i made some change to this code
I change the torch.backends.cudnn.benchmark = False for a train error
I set the cuda devices = 1
i dont know weather the test problem is caused by these

[ZhenLiuBuaa]

i find in test process ,kp.py line 265, kp = kp_(inter)
kp data is unnormal ,some value is bigger than 1e30 and inter data is right

[ZhenLiuBuaa]

this is kp data @Duankaiwen
kp.txt

_

[ZhenLiuBuaa]

this is the test error
Traceback (most recent call last):
File "test.py", line 99, in
test(testing_db, args.split, args.testiter, args.debug, args.suffix)
File "test.py", line 61, in test
testing(db, nnet, result_dir, debug=debug)
File "/home/lz/python/CenterNet/CenterNet-master/test/coco.py", line 321, in testing
return globals()[system_configs.sampling_function](db, nnet, result_dir, debug=debug)
File "/home/lz/python/CenterNet/CenterNet-master/test/coco.py", line 317, in kp_detection
db.evaluate(result_json, cls_ids, image_ids)
File "/home/lz/python/CenterNet/CenterNet-master/db/coco.py", line 176, in evaluate
coco_dets = coco.loadRes(result_json)
File "data/coco/PythonAPI/pycocotools/coco.py", line 318, in loadRes
if 'caption' in anns[0]:
IndexError: list index out of range

[Duankaiwen]

@ZhenLiuBuaa Sorry, i'm a little busy these days, so i did't reply to you immediately. Please put 'pdb.set_trace()' in line 93 in test/coco.py to make sure the images are read correctly.

[ZhenLiuBuaa]

Thanks for your reply.I think maybe the cudnn.benchmark value influence the model structure. The saved model is a little bigger than what you provided. Good luck to you. 发件人: Kaiwen Duan 发送时间: 2019年6月10日 20:19 收件人: Duankaiwen/CenterNet 抄送: ZhenLiuBuaa; Mention 主题: Re: [Duankaiwen/CenterNet] a test problem (#50) @ZhenLiuBuaa Sorry, i'm a little busy these days, so i did't reply to you immediately. Please put 'pdb.set_trace()' in line 93 in test/coco.py to make sure the images are read correctly. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[yushanshan05]

@ZhenLiuBuaa I met the same problem. Can you telll me how to solve this problem. Thanks!

[ZhenLiuBuaa]

@ZhenLiuBuaa I met the same problem. Can you telll me how to solve this problem. Thanks!

i think if you want test the model,you should set cudnn.benchmark=True.if not ,you can train your model with cudnn.benchmark=False