maobj

Thank you for your quick reply, I check the code but find train_aug=False. Maybe not this reason 

I tried the pre-train model in FEAT, but I find worse result Test Acc **0.5760 + 0.0082** in 1 shot and **0.7381+0.0066** in 5 shot. Could you upload your pre-trained...

> I feel strange, I am at 512 resolution, single card bs=8. 当我把"sample_n_frames: 16mixed_precision_training" and "enable_xformers_memory_efficient_attention"设置为True后，可以正常跑起来。但是还有个问题，在train_hack.py里设置poseguider的输出为320维（poseguider = PoseGuider(noise_latent_channels=320)），但是在推理的时候python3 -m pipelines.animation_stage_1 --config configs/prompts/animation_stage_1.yaml，加载poseguider模型时，设置的channel是4，model = PoseGuider(noise_latent_channels=4)，而且在后续的pipeline 里面使用的方式对应的也是channel为4的情况（latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) + latents_pose）。这个是为什么呢？好像是推理的时候没有用你的hackunet结构？

Your generated results are impressive, particularly in the face region. Could you share your techniques for improving the quality of the facial area? In our experiment, when the input size...