CoordConv Implementation with Pytorch

• Unofficial Pytorch implementation of the paper An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution

0. Develop Environment

Docker Image
- pytorch/pytorch:1.8.1-cuda11.1-cudnn8-devel

• Using Single GPU

1. Implementation Details

• data : generate regression dataset
• dataset.py : regression dataset
• main.py : run train with argparse
• model.py : regression models used in paper
• train.py : train, test, inference model
• Check AddCoord.ipynb : check CoordConv
• Visualize - Figures (CoordConv).ipynb : visualize 100 figures from CoordConv models per hyperparameter settings
• Details
  • Official codes are weird cause there are no activation functions on CoordConv model
  • No learning rate scheduler for convenience

2. Reference

• An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution [paper] [official code]