scnuhealthy/animate

stars

commits

Python

primary language

Jul 24, 2024

updated

README

关于 do_classifier_free_guidance得实现，首先基于一个思路，讲latent (b,c,h,w)复制为(2b,c,h,w),前面一半是uncond,后面一半是正常得cond

reference_control_reader: uncond得latent沿用self-attention（不加入ref特征）,然后cond得latent用animateanyone得实现，即合拼referencenet的特征一起做self-attn
ref_image_latents，latent_pose 复制
noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
这里chunk是将(2b,c,h,w)拆为两个(b,c,h,w)，分别代表uncond和cond

17 commits

lucidrains/classifier-free-guidance-pytorch

Implementation of Classifier Free Guidance in Pytorch, with emphasis on text conditioning, and…

544

shiyi-zh0408/FlexiAct

[SIGGRAPH 2025] Official code of the paper "FlexiAct: Towards Flexible Action Control in…

341

samuruph/Ctrl-Adapter

alibaba/animate-anything

Fine-Grained Open Domain Image Animation with Motion Guidance

971

HL-hanlin/Ctrl-Adapter

Official implementation of Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse…

474

benjamin-paine/dragnuwa-pruned-safetensors

JiauZhang/Text2Video-Zero

Implementation of Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators

DREAMXFAR/Shape-aware-ControlNet

It is the official implementation of Shape-aware ControlNet, accepted by ACM MM'2024.

Python

100.0%

README

关于 do_classifier_free_guidance得实现，首先基于一个思路，讲latent (b,c,h,w)复制为(2b,c,h,w),前面一半是uncond,后面一半是正常得cond

reference_control_reader: uncond得latent沿用self-attention（不加入ref特征）,然后cond得latent用animateanyone得实现，即合拼referencenet的特征一起做self-attn
ref_image_latents，latent_pose 复制
noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
这里chunk是将(2b,c,h,w)拆为两个(b,c,h,w)，分别代表uncond和cond