Clothes Detector Model

Convolutional Neural Network Model detecting clothes

Architecture

Faster R-CNN with ResNet+FPN backbone

Literature

• Faster R-CNN https://arxiv.org/pdf/1506.01497.pdf
• ResNet https://arxiv.org/pdf/1512.03385.pdf
• FPN https://arxiv.org/pdf/1612.03144.pdf

Setup

Clone repository

git clone https://github.com/pawelbeza/ClothesDetectorModel.git

Create conda environment

conda env create --file env.yml
conda activate clothes-detection

Prepare dataset

1. Download DeepFashion2 dataset to dataset dir from google drive
2. Create coco annotations

mkdir dataset/detectron_annos  
python3 dataset/deepfashion2_to_coco.py --out_dir ./dataset/detectron_annos/train_annos_all.json --image_dir ./dataset/train/image --annotation_dir ./dataset/train/annos --num_images 191961  
python3 dataset/deepfashion2_to_coco.py --out_dir ./dataset/detectron_annos/validation_annos_all.json --image_dir ./dataset/validation/image --annotation_dir ./dataset/validation/annos --num_images 32153

Training

python3 train_net.py --config configs/Faster_RCNN_R101_FPN_3x.yaml --resume