DeepStyle: Using State-of-the-Art Deep Learning to Generate Realistic High Fashion Clothing and Styles
This is the official repository for the Medium Article Part 1 and Part 2 that contains all the relevant source code, jupyter notebooks, and configuration files for the paper. The paper is also included in the repository and can be found here.
The .yaml files contain the configuration details of the object detection models trained in this research as described in the paper.
The DeepFashion2Coco.ipynb shows the steps to convert the DeepFashion Dataset to COCO format to ease the training and validation of the models implemented by the Detectron framework.
The vis.py file is under the detectron/utils/ folder in the Detectron github repository. Replace the vis.py file in the repository with the updated one to reproduce the cropping of the test images based on the predicted bounding boxes, as described in the paper. This vis.py adds the code for detecting full-body fashion items in images, cropping them based on the bounding boxes, and saving the results to a local directory.
Finally, the rest of the jupyter notebooks provide the implementation details of the DCGANs as described in paper. The hyperparameters for training and the architectures for the generator and discriminator are provided for each model tested. Code are written in Pytorch.