KeqiangSun
KeqiangSun
> i observed the same problem.. > do you have any solution? Adding `torch.cuda.set_device(rank)` in the beginning of the training function might help.
> 我理解的论文中所讲，是每种舞蹈的动作的采样频率都不一样，不是25FPS的固定值。 > 所以论文中有讲到alignment的问题。 > 我觉得config.json中start和end都是有用的，因为需要根据总的帧数对动作进行插值？ 我们根据这样的方法进行了重新对其，发现动作和音乐还是对不上。请问有人成功使用这个数据吗？希望能给出详细些的说明，感谢！
Thanks for your contribution. but there seems to lack the weight named 'conv4_2' and 'fc7/W' when I try to finetune the pretrained VGG16 model. could you please tell me why...