https://github.com/mmz33/capsnet
A Tensorflow Implementation of Capsule Network
https://github.com/mmz33/capsnet
capsule-network cnn computer-vision neural-networks tensorflow
Last synced: about 2 months ago
JSON representation
A Tensorflow Implementation of Capsule Network
- Host: GitHub
- URL: https://github.com/mmz33/capsnet
- Owner: mmz33
- Created: 2018-10-12T08:47:07.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2020-02-22T22:33:15.000Z (over 6 years ago)
- Last Synced: 2025-08-25T17:28:12.528Z (10 months ago)
- Topics: capsule-network, cnn, computer-vision, neural-networks, tensorflow
- Language: Python
- Homepage: https://mmz33.github.io/Capsule-Networks/
- Size: 109 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Capsule Network
This repository represents a TensorFlow implementation of capsule network (CapsNet). For more details about CapsNet, you can check my [blog post](https://mmz33.github.io/Capsule-Networks/). you can also find [here](https://github.com/mmz33/CapsNet/blob/master/report/capsnet_seminar_report.pdf) a detailed report about this topic.
## Files
- `dataset.py`: loads the MNIST dataset using keras API
- `capsnet.py`: represents CapsNet architecture and contains function to build it
- `capsule_layer.py`: represents CapsNet layers which are mainly PrimaryCaps and DigitCaps layers
- `engine.py`: it extracts parameters from the config, set up training and testing configuration, and implements them
- `config.py`: represents a dict of parameters with a getter function
- `main.py`: the main entry point
- `utils.py`: contians some helping functions
- `run_kaggle.py`: a script to run digit recognizer competition from kaggle
## Training
For training, run `python3 main.py --train`. In `config.py`, you can specify your hyperparameters. `checkpoint_path` is the location where models/checkpoints are saved. `log` is the location where TensorFlow summaries are saved to be used later in Tensorboard for example.
## Testing
For testing, you just need to run `python3 main.py --test`. This will load the model corresponding to the latest saved checkpoint.
#### Kaggle Digit Recognizer
In addition, the code was tested on the test data provided in digit recognizer competition from Kaggle, which is also MNIST data. The score achieved was: `0.99500` which is `99.5%` accuracy.
## Tensorboard
#### Train
#### Reconstructed Images
##### During training
##### During validation
#### TF Graph
