CatImageGeneration

Prerequisites

• Python 3
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Clone this repo
• Create env conda create --name pix2pix
• Install requirements pip install -r requirements.txt

train/test

• To download pretrained model and dataset run bash bin/download.sh

• Alternative way to download via GD link link

• Train a model: python train.py --train_dir ../data_examples/train/ --device cuda --checkpoint_dir ../models/ You can also set lr, max_epochs, l1_coef, checkpoint_freq, and val_dir

• Generate: python generate.py --edge_dir .../data_examples/generate/ --device cuda --gen_path PATH_TO_PRETRAINED_MODEL --save_dir ../data_examples/generated_imgs/

Results

The model was trained on the edge2cats dataset, edges were obtained by pretrained DexiNed link, cats faces dataset link. Also model was trained on the maps dataset link.

1st col: Input / 2nd col: Generated / 3rd col: Target