Yuhao Xu

pip install opencv-python

Our gradio demo requires 2 GPUs. Since ootd_dc checkpoints are not released yet, you may simply comment out the code involving ootd_dc for now.

> > Our gradio demo requires 2 GPUs. Since ootd_dc checkpoints are not released yet, you may simply comment the code involving ootd_dc for now. > > Any ETA for...

model_path & cloth_path should be a specific image path like "./examples/model/xxx.png"

>  Am I doing this right? I don't know which model path to use. model path means the target model (person) image path.

Hi! Thanks for your question. It is because the clothing features are learned in the latent space rather than being simply pasted. Please try larger guidance scales and different random...

Thanks for your question. Both outfitting and denoising UNets are finetuned on the original SD1.5 UNet. The training details are mentioned in our paper, which will be released soon. Please...

> > Thanks for your question. Both outfitting and denoising UNets are finetuned on the original SD1.5 UNet. The training details are mentioned in our paper, which will be released...

我们这里没有对prompt embeds做classifier-free guidance，而是对garment latents做，可以看一下这个函数https://github.com/levihsu/OOTDiffusion/blob/1347a6b88118d7c508966d6e59dd641a236ded86/ootd/pipelines_ootd/pipeline_ootd.py#L701 论文这周内就会放出来，具体细节可以稍等一下哈

https://github.com/levihsu/OOTDiffusion/issues/59#issuecomment-1965683568 可以试一下这个方法 `os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128"`