Got about zero acc When validating resnet50_no_fusion

[TimeEternal]

Training log is like below, train_acc is around 1, but test_acc is 0, only used h36m dataset. I also train the fusion model, but i got same train_acc and test acc What happened? Is the model overfitting? 256_nofusion_2019-12-10-09-57_train.log

[CHUNYUWANG]

Can you help check if the saved debug images in the "output" directory are correct? In particular, the images with names "validation_".

[CHUNYUWANG]

I tried to run the noFusion experiment. The log of the first epoch is as follows:

Test: [0/1250] Time 26.000 (26.000) Loss 0.0133 (0.0133) Accuracy 1.000 (1.000) Test: [100/1250] Time 0.212 (0.411) Loss 0.0334 (0.0276) Accuracy 0.971 (0.981) Test: [200/1250] Time 0.102 (0.274) Loss 0.0252 (0.0288) Accuracy 0.996 (0.981) Test: [300/1250] Time 0.107 (0.225) Loss 0.0325 (0.0298) Accuracy 0.978 (0.978) Test: [400/1250] Time 0.151 (0.201) Loss 0.0421 (0.0301) Accuracy 0.945 (0.976) Test: [500/1250] Time 0.158 (0.187) Loss 0.0127 (0.0310) Accuracy 1.000 (0.973) Test: [600/1250] Time 0.155 (0.178) Loss 0.0169 (0.0298) Accuracy 1.000 (0.975) Test: [700/1250] Time 0.112 (0.170) Loss 0.0309 (0.0293) Accuracy 0.985 (0.976) Test: [800/1250] Time 0.158 (0.165) Loss 0.0288 (0.0291) Accuracy 0.985 (0.976) Test: [900/1250] Time 0.092 (0.162) Loss 0.0306 (0.0291) Accuracy 0.989 (0.976) Test: [1000/1250] Time 0.143 (0.159) Loss 0.0290 (0.0294) Accuracy 0.978 (0.976) Test: [1100/1250] Time 0.156 (0.156) Loss 0.0329 (0.0292) Accuracy 0.985 (0.976) Test: [1200/1250] Time 0.119 (0.154) Loss 0.0250 (0.0304) Accuracy 1.000 (0.972)	Arch	root	rhip	rkne	rank	lhip	lkne	lank	belly	neck	nose	head	lsho	lelb	lwri	rsho	relb	rwri
256x256_multiview_pose_resnet_50_d256d256d256	1.000	0.983	0.966	0.910	0.987	0.947	0.686	0.990	1.000	1.000	1.000	0.917	0.816	0.776	0.928	0.797	0.725

[TimeEternal]

I reclone the respository， it works properly now. Thanks for your reply