1. Introduction
2. Installation
3. Preparedata
4. Train
5. Tests
6. Datasets

The pattern_replacement program replace input cloth with given pattern based on cloth trend. This process can remove original pattern on cloth.

Requirements shortlists:

• Pattern replacement program
• Prepare data

Run demo on given data requires:

• NVIDIA GPU, Python3
• Tensorflow-gpu
• various standard Python packages in requirement.txt
• Pillow

Notes:

• The program was tested on tensorflow-gpu 2.3.1 with cuda 10.1 and cudnn 7.6
• The program requires your opencv-contrib-python and opencv-python in the same version

Prepare your own data:

To run you own data on this demo, you will need to prepare the Segmentation result of image, and run Densepose to get the human body trend information.

The main requirement of these programs:

• Caffe2,COCOAPI
• PyTorch >= 1.4.0

The details requirement of Segmentation and Densepose can also be found in the link.

1. Put input image under data/input/
2. Run segementation program and put result under data/seg/
3. Run

python preparedata.py

The result will under data/mask/

4. Run densepose and put result under data/IUV/

After done the requirement prepare work:

Clone repository:

git clone https://github.com/aircat1216/fashion_pattern_replacement your_path

Install Python dependencies:

pip install -r your_path/requirements.txt

If you want to run given demo, just run:

python predict.py

And you will find gray_scale result in data/output_gray/ and finan result in data/output_final.

To use your own data:

1. Put the input images under data/input/
2. preparedata and put mask under data/mask, IUV images under data/images
3. run

python predict.py

If you want to select your own model and pattern, run:

python predict.py your_model_weight_file_name.h5 your_pattern_path

See Preparedata to prepare your own data