kulasama / deepfashion

DEEP FASHION

Setup Environment

# Virtual environment (optional)

# Tensorflow (optional)
sudo apt-get install python-dev libhdf5-dev libblas-dev  liblapack-dev
conda create -n deepfashion python=3.7
source activate deepfashion
conda install tensorflow-gpu keras  # for Python 2.7 and GPU

# Dependencies
sudo apt install -y python-tk
pip install -r requirements.txt 

Download DeepFashion Dataset

# http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion/AttributePrediction.html
./dataset_download.sh

# The directory structure after downloading and extracting dataset:
# fashion_data/
# ---Anno
# ------list_attr_cloth.txt
# ------list_attr_img.txt
# ------list_bbox.txt
# ------list_category_cloth.txt
# ------list_category_img.txt
# ------list_landmarks.txt
# ---Eval
# ------list_eval_partition.txt
# ---Img
# ------img

Create Dataset

# For images in fashion_data, apply selective search algo to find ROI/bounding boxes. Crop and copy these ROI inside dataset
python dataset_create.py

Train

python train.py

Predict

python predict.py

Misc

dataset - Contains images used for training, validation and testing.

output - Contains trained weights and bottleneck features.

logs - Contains logs and events used by tensorboard.

MODEL

						->	Classification Head (Categories)
InputImage	->	VGG16 + Layers	--
						->	Regression Head	(Confidnence in the Classification head prediction)

RESULTS

Acknowledgment

• DeepFashion Dataset