mutoy-choi / tryondiffusion

Tryondiffusion

Introduction

This repository is an attempt to implement the Tryondiffusion model. For more details, visit the official Tryondiffusion website.

Environment

The code was developed and tested on the following environment:

• Operating System: Windows server 2019

• Python Version: Python 3.10

• GPU : NVIDIA Tesla T4 16GB

• CUDA 11.8

Getting Started

Clone the Repository

To get started with training examples, first clone this repository by running the following command in your terminal:

git clone https://github.com/Mutoy-choi/Tryondiffusion


cd Tryondiffusion

This will clone the repository and navigate you into the project directory.

Set Up Virtual Environment

python -m venv venv

.\venv\Scripts\activate

These commands create and activate a virtual environment named venv. This isolates the project dependencies, making it easier to manage.

Install Dependencies

pip install -r requirements.txt

Run the Example Training code

python one_shot_test_ParallelUnet.py

This file allows you to know if the model is working well or not using example data for ParallelUnet(From 128x128 to 256*256)

Further upload

• update preprocessing AI-Hub data

https://aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=78

Deployment content and amount of data provided

Studio fashion video (model photo) 6,741,328 cases

Studio Fashion Video Model Key Points: 120,936 Cases

Studio Fashion Video Model Semantic: 120,936 cases

Fashion products and fashion video pair: 117,270 cases

Fashion product representative photos (product photos) 40,036 cases

Fashion product key points: 40,036 cases

Fashion Product Semantic: 40,036 cases

tryondiffusion's Issues

After I trained my weights, how can I inference with weights?

Excuse me, how to generate landmarks in Jp and Jg ？

Excuse me, how to generate landmarks in Jp and Jg ？

Hello! How is it going? Need help?

Hello! How is it going (are weights available)? Need help? if Yes - please contact via telegram: dl_hello

no module named CustomDataset

concatenate the segmented garment Ic and the noisy image zt

In paper, 'concatenate the segmented garment Ic and the noisy image zt along the channel dimension' ,in this repo,concatenate two pictures along the H dimension,Do I have an exact understanding?

The preprocessing AI-Hub data is difficult to download.

Thank you for your enthusiasm and effort in implementing the code. I would like to know if there is any other way to download preprocessing AI-Hub data, such as Google Drive.

There is some mistakes in "UNetBlock" function

Tryondiffusion/parallelUNet128.py

Line 75 in 9508623

def forward(self, x, garment_features=None, pose_embedding=None):

Thanks for sharing code! When I run the code, I find some mistakes:

1. when in_channels != out_channels, will go errors. This can be fixed by modify line 67-68
2. TypeError: forward() got an unexpected keyword argument 'garment_pose_embedding'

diffusion process

hi! Thanks for your great coding work. However, I cannot find the diffusion process in your code, it seems there is no diffusion forward-reverse process right?

”Output1_upsampled“ is not input to model2.

The 256x256 image is not input to model2. Should model2 be the 256x256 model? And should the param in "parallel_config_256" be changed?