Weixia Zhang
Weixia Zhang
@tstandley Thanks for the reply, it's very helpful! I noticed that in Line 173, a mak is generated, where the original values less than 2^13 are set to be valid....
@hadesfgh 你好，生成失真图片的代码只有基于Matlab的，也不准备将它迁移到python上了，另外这个项目是几年前做的，预训练的图片电脑上已经没有保存，不好意思。 生成图片的代码应该并不难理解，可能需要在"distorted_img = distortion_generator( img, dist_type, level, seed )"外层写三for循环分别遍历原图(img)、失真类型(dist_type)和失真等级(level)，并保存在相应的文件夹中即可。
不好意思，这个项目我们没有实现过多gpu
https://huggingface.co/datasets/teowu/LSVQ-videos
@michelm117 Hi, did you find the mesh_names.json files finally? I am also confused with this problem.
Yes, I am using master pytorch
I solved this problem by adding extra_compile_args = ['-std=c99'] into the ffi, but I am still curious about the reason for this....
@Redaimao See the file of roi_align as an example, which has already added this arg into cffi.
Maybe, try using smaller batch size
You may also try setting opt = 1 in Line127, which freezes the weights of text encoder. Empirically, this would not affect the final performance very much, but can significantly...