papa issues

Results 5 issues of


                                            papa

想请教一下论文中的3D full attention的实现具体在哪里呢？

看到paper中提出了3D full attention和2D+1D attention的对比，并实验证明说生成视频的效果更好。论文中也提到可以用多种并行策略优化，想请教一下这部分有开源的代码吗

Support for multi-node training or inference？

Does the worker process in the ProcessWorkerWrapper class only support single-node parallel work?

How consistent self-atttention fits into the semantic motion predictor？

Great work by everyone! I'd like to ask you a little bit about how consistent self-atttention fits into the semantic motion predictor, I see that the input in the semantic...

Where is the concrete implementation of consistent self-atttention?

Great work by everyone! I would like to ask for leave. Where is the concrete implementation of consistent self-atttention?

Would you mind asking if there is any conflict between the Multi-Resolution Frame Pack method you mentioned in the paper and the progressive training method?

Progressive training is divided into different stages. For instance, in the first stage, the resolution is 256px. Would the video data from different resolutions and frames still be mixed?