wang yilan

先感谢一下辛苦的开发人员~ 我想在自己的视频上跑一下nextvlad，但是我发现这个需要读取一个.pickle文件，这个文件应该是百度自己把youtube8m测试集转码的格式吧？ 想问问如果想在自己的视频上跑一下，我要怎么做呢？（版本，code什么的都已经ready了）

Excuse me, why can I only run one img on one GPU?

14

Why can only one picture run on one gpu now? If I want to modify, from what code and where should I modify it? Many thanks！

I train this great work, but I find the perceptual loss is so big, e.g, the L1loss is 10e-1, and the perceptual loss is 10e+1……Dose I got some errors? thx!

question of ADD(Adversarial Diffusion Distillation), the inputs of teacher nets

4

in ADD, why the inputs of teacher nets are the denoised results, rather than the same noise inputs of student nets? many thanks！

NotImplementedError: Graph op: /quantize/ArgMin(ArgMin) has no backend implementation

2

RuntimeError: Op Execution Error: /quantize/ArgMin(Type: ArgMin, Num of Input: 1, Num of Output: 1) 请问我们有计划支持argmin/argmax吗

请问model zoo的模型在哪里？

1

如题，比如分类任务里， MODEL_DIR = 'QuantZoo/Model/Imagenet' 请问这些模型是不是需要我去torchvision下载，然后自己转成onnx，然后再来测试呢？我们有解决这种需求的脚本吗？感谢~~

How to generate HQ dataset?

2

thanks for this great job! The resolution of FFHQ original dataset is 1024x1024, but in your paper the resolution of HQ data is 512x512. So how to generate the 512x512...

Is it normal for big loss in stage2, cross entropy loss?

6

I find the cross_entropy_loss is big in training stage2, between 1e-1~1e0. From a numerical perspective, it seems that this represents almost no optimization of codebook searching.

cannot install channelnorm-cuda

1

in pypi or conda, there are none of channelnorm-cuda. Could you tell me how to install it? thx!

very thanks for your job! and I still have a little question, in fakd, even if the feat map size is [256x256], the size of affinity matrix is [1,1,65536,65536], and...