Max Shek-wai Chu

Thanks for the reply, would you consider adding this feature?

@sunke123 First of all, MPII released the test set ground truth. So you can download it from http://human-pose.mpi-inf.mpg.de/#download. Also,, if you want to use the pred.mat generated by this HRNet...

> 1、could you please tell me the way to set "test_scale" is test_scale=[1,2,3,4,5,6] right ? > but `RuntimeError: invalid argument 2: input and output sizes should be greater than 0,...

@touhou-ayaya Have you downloaded the evaluation script from MPII (http://human-pose.mpi-inf.mpg.de/results/mpii_human_pose/evalMPII.zip)? After downloading and unzip it, copy my updated evaluation scrip above and you can directly evaluate MPII accuracy from the...

> @Sunstin > I run the code but I cannot get the results of 91.8. Could you tell me how you do the testing more detailly? What accuracy did you...

@Sunstin that means your result is using only the flip test. You can use the multi-scale code i post to obtain the multi-scale test, which should give your 91.8%

@XiyueSun my multi-scale should reproduce successfully the validation accuracy. However, it cannot reproduce the test accuracy.

> @MaxChu719 Hello,why i can't find test set ground truth on http://human-pose.mpi-inf.mpg.de/#download? Please help me! it seems they removed it. I am uploading the ground truth file to my google...

> > Same problem. > > In test.json provided by HRnet, the test.json size is 7247. > > But in https://github.com/anibali/eval-mpii-pose, the test.h5 size is 11731. > > Also, the...

``` [Chus-MacBook-Pro:deeppose chumax$ bash shells/train_lsp.sh 2016-09-20 11:08:00,279 [INFO] sys.version_info(major=2, minor=7, micro=12, releaselevel='final', serial=0) 2016-09-20 11:08:00,279 [INFO] chainer version: 1.15.0.1 2016-09-20 11:08:00,279 [INFO] cuda: False, cudnn: False 2016-09-20 11:08:00,279 [INFO] Namespace(adam_alpha=0.001,...