Cartoon-entity-extraction

Run for Garfield comics dataset. Preprocessing involves using PixelAnnotation tool to create training masks. Unet (Pytorch) was run to segment the data on an NVIDIA RTX2070 8GB GPU

Installations:

• Python 3.6
• pip install requirements.txt
• CUDA 10.2, cuDNN
• Download checkpoint from here.

References:

• Paper: UNET
• Github: milesial/Pytorch-UNet

Dataset:

• Annotated using abreheret/PixelAnnotationTool which can be downloaded by doing the following:

  wget https://github.com/abreheret/PixelAnnotationTool/releases/download/v1.3.1/PixelAnnotationTool_x86_64_v1.3.1.AppImage
  chmod +x PixelAnnotationTool_x86_64_v1.3.1.AppImage
  ./PixelAnnotationTool_x86_64_v1.3.1.AppImage
  

• For training: Under the data folder
• For testing validation: Can be parsed and generated. Used the dataset from here
• 

Training:

• Edit data loader and filepaths and run python3 train.py

Running Inference:

src = '/path/to/test_images'
tgt = '/path/to/saved_images'
for image in os.listdir(src): 
    os.system("python predict.py -i " + src + image + " -o " + tgt + image + " --model checkpoints/CP_epoch45.pth") 

Results: