aguzel / ChromaFashionette

Designing Efficient Image-to-Image Translation Artificial Neural Network Model For Segmenting Fashion Images

DOWNLOAD PAPER

ChromaFashio[net]te 👗 🧠

Designing Efficient Image-to-Image Translation Artificial Neural Network Model For Segmenting Fashion Images

ChromaFashionette is a name for the artificial neural network built during my University of Leeds AI MSc thesis. It combines the word "chroma," which refers to color, with " fashio[net] te," which is fitting for a network that segments fashion images, and the ending "-ette" that denotes smallness or subtlety (efficient and small).

Install dependencies

pip3 install -r requirements.txt

Dataset

--data
  -train
   * A
   * B
  -test
   * A
   * B

Run training

python3 train.py
python3 train_weighted.py

Run testing

python3 test.py

Training Settings

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
BATCH_SIZE = 4
NORMALIZE = False
ARCHITECTURE = 'DeepLabV3+'
NUM_CLASSES = 5 
LR = 1e-4
EPOCHS = 5