Doodle Classifier & Generator for QuickDraw!

CDCGAN Generator and ResNet34 Classifier for QuickDraw! dataset from Google

Demo

airplane	bicycle	butterfly	cake	camera

chair	clock	diamond	The_Effiel_Tower	tree

Doodle Classifier

Model: ResNet34

	Train	Test
Loss
Accuracy	99%	96%

• Prepare training data

  cd Classification
  python download_data.py -c categories.txt -r Data
  python ./DataUtils/prepare_data.py -root Data -msc 10000 -v 0.2

• Start Training

  python Classifier.py -e 40 -bs 64 -lr 0.1 -m resnet34

• Evaluation

  python Evaluation.py -i ***.npy
  

Generator

Model: DCGAN / DCCGAN

	Discriminator Loss	Generator Loss	Result
airplane
camera

• Prepare training data

  cd Generation
  python download_data.py -c categories.txt -r Data

• Start Training
  1. DCGAN

     python dcgan.py -o airplane -e 40 -log 1 -lr 5e-5

  2. DCCGAN

     python dccgan.py -c 30 -s 50000 -e 4 -log 1 -bs 64

• Evaluation
  1. DCGAN

     python Evaluation.py -r models/airplane -m DCGAN

  2. DCCGAN

     python Evaluation.py -r Trained_models -m DCCGAN -c 30

Credit

Thanks these guys.

• XJay18/QuickDraw-pytorch
• eriklindernoren/PyTorch-GAN
• togheppi/cDCGAN