Skip to content

Latest commit

 

History

History
28 lines (20 loc) · 1.11 KB

README.md

File metadata and controls

28 lines (20 loc) · 1.11 KB

DCGAN - Deep Convolutional Generative Adversarial Networks

This project is conducted for the sake of re-implementation of the main paper. The project is implemented mainly with Pytorch.
Paper: Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

Data

LSUN

The paper used the LSUN which is a Large-scale Image Dataset with approax 3 milion bedroom scence images.

Traning

  • Modify config file in /src/cfg to setup for training process or
  • Set parameters directly with dcgan_lsun.py

Run

python3 src/models/dcgan_lsun.py

Result

Generation over interpolated point in noise dimension

Generation over interpolated points between 11 random noise code to see transition between points.

Guided backpropagation to visualize learned feature of Discriminator

Last row illustrates reconstructed images from learned features of trained Discriminator. Middle row is images which conversely are attained from a non-trained Discriminator.