linghai06

To be more concrete, modified codes and training scripts are attached as follows:  > self.train, self.valid = self.train[:4], self.valid[:4] is added in datasets/mpii.py to select only 4 images And...

Well, literally it is not training in normal sense but testing if the gradient flow could drive the network to memorize such a simple setting. If not, then probably there...

Only 4 images were used and the batch-size was set to 4. Currently I only pay attention to the training loss for this 4 images and surprisingly found that it...

> Thanks for the sharing~ > I also met the problem. I trained 30k person images and generated 2 black images. I found the loss is NAN. >  you...

@MSBlueVn did you solve the problem? i have the same issue

@dongzhuoyao no, they are not. check this: