text2painting

Convert keywords into painting

Sample results (input words - stage1 output - stage2 output - stage3 output - ground truth)

Model

Model is built as sequential GANs with three stages

Stage-1 structure

Stage-2 structure

Stage-3 structure

Installation

Set environment with

conda env create -f environment.yml

Install Spacy English language model

python -m spacy download en

Demo

Visit https://painter-ai.com .

Data

You can download dataset from

A small sample of dataset can be downloaded from

https://drive.google.com/open?id=1KXEIrRGDvASEEm-vT_jJcvFLncapJDEL

Place them under data/deviantart_verified/images and data/wikiart_verified/images

Model

There are models in the same link above as well

https://drive.google.com/open?id=1KXEIrRGDvASEEm-vT_jJcvFLncapJDEL

Place a model directory under models/

Usage

All the necessary parameters are in lib/config.py

Train

python train.py

Continue training with

python train.py --model <model_file>

Test

python test.py --model <model_file>

Prediction

python pred.py --model <model_file> --input <input_text_file>

Example input text file format is asset/input.txt with each keyword set in a newline.

Citation

Please cite this paper

https://arxiv.org/abs/2007.04383