https://github.com/renan-siqueira/autoencoder-research
A comprehensive toolkit built with PyTorch, designed to facilitate the training, evaluation, and visualization of autoencoders. From simple linear autoencoders to convolutional and variational architectures, this project offers an intuitive and expandable framework for anyone delving into the realm of unsupervised learning.
https://github.com/renan-siqueira/autoencoder-research
autoencoder computer-vision container convolutional-neural-network data-augmentation deep-learning dimensionality-reduction docker dockerfile image-processing image-reconstruction machine-learning model-checkpoint neural-networks python pytorch unsupervised-learning variational-autoencoder virtualization
Last synced: about 2 months ago
JSON representation
A comprehensive toolkit built with PyTorch, designed to facilitate the training, evaluation, and visualization of autoencoders. From simple linear autoencoders to convolutional and variational architectures, this project offers an intuitive and expandable framework for anyone delving into the realm of unsupervised learning.
- Host: GitHub
- URL: https://github.com/renan-siqueira/autoencoder-research
- Owner: renan-siqueira
- License: mit
- Created: 2023-10-13T18:45:20.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2023-10-16T22:02:59.000Z (over 2 years ago)
- Last Synced: 2023-10-17T15:38:27.743Z (over 2 years ago)
- Topics: autoencoder, computer-vision, container, convolutional-neural-network, data-augmentation, deep-learning, dimensionality-reduction, docker, dockerfile, image-processing, image-reconstruction, machine-learning, model-checkpoint, neural-networks, python, pytorch, unsupervised-learning, variational-autoencoder, virtualization
- Language: Python
- Homepage:
- Size: 44.9 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 3
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Autoencoder Project
A simple implementation of an autoencoder using PyTorch.
This project aims to provide a basic framework for understanding, training, and evaluating autoencoders on any image size.
## Features
- Train various autoencoder architectures: standard, convolutional, variational, and convolutional variational autoencoder on your dataset of images.
- Visualize the reconstructions of the autoencoder.
- Evaluate the model on a separate validation set.
- Checkpointing: Ability to save and resume training from checkpoints.
- Save and load trained model weights.
- Utilize custom datasets by simply pointing to your directory.
- Test all available autoencoder architectures with a single command.
- Easily containerize and run using Docker.
## Getting Started
### Prerequisites
- Python 3.x
- PyTorch
- torchvision
- PIL
- matplotlib
- NVIDIA CUDA-compatible GPU (if you're using the GPU-support Docker setup)
- Docker (if you want to containerize the application)
### Installation
#### Traditional Setup:
1. Clone the repository.
2. Navigate to the project directory and install the required libraries.
#### Using Docker:
1. Clone the repository.
2. Navigate to the project directory.
3. Build the Docker image:
```bash
docker build -t autoencoder_research .
```
4. Run the container:
```bash
docker run --gpus all -it --rm -v $(pwd):/app autoencoder_research bash
```
**Note:** If you're using the Docker method, you'll be inside the container's shell after running the above command. You can execute Python scripts or any other commands just like you would on your local machine.
## Usage
1. Modify the `settings/settings.py` file to point to your training and validation dataset.
2. Modify the `json/params.json` file to reflect your training preferences.
### Training
To train the autoencoder as per your configurations, simply run:
```bash
python3 run.py
```
### Testing
To run a test routine for all the available autoencoder architectures, use:
```bash
python3 run.py --test
```
This will train each autoencoder for a few epochs and provide an overview of their performances.
## Common Issues and Solutions
### Issue: Docker Volume and Uppercase Paths on Windows
**Description:** On Windows, when trying to mount a local volume in Docker, an error might arise if the path contains uppercase letters. This is due to Docker expecting repository names (or paths) to be in lowercase.
**Solution:** Convert the current directory path to lowercase and replace backslashes with regular slashes. If you're using Git Bash or a similar terminal on Windows, follow the steps below:
1. Get the full path of the current directory in Windows format:
```bash
win_path=$(pwd -W)
```
2. Convert the path to lowercase and replace backslashes with regular slashes:
```bash
lowercase_path=$(echo $win_path | tr '[:upper:]' '[:lower:]' | sed 's|\\|/|g')
```
3. Use this path in your Docker command:
```bash
docker run --gpus all -it --rm -v "/$lowercase_path":/app autoencoder_research bash
```