https://github.com/soumik12345/enet

Pytorch Implementation of ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation (https://arxiv.org/abs/1606.02147)
https://github.com/soumik12345/enet

camvid-dataset deep-learning enet mish pytorch real-time-semantic-segmentation semantic-segmentation

Last synced: 27 days ago
JSON representation

Pytorch Implementation of ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation (https://arxiv.org/abs/1606.02147)

• Host: GitHub
• URL: https://github.com/soumik12345/enet
• Owner: soumik12345
• Created: 2019-12-09T17:01:53.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-01-24T10:45:17.000Z (over 5 years ago)
• Last Synced: 2025-05-12T13:12:22.280Z (27 days ago)
• Topics: camvid-dataset, deep-learning, enet, mish, pytorch, real-time-semantic-segmentation, semantic-segmentation
• Language: Jupyter Notebook
• Size: 47.6 MB
• Stars: 11
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Enet

[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/soumik12345/Enet/master)

[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/enet-a-deep-neural-network-architecture-for/real-time-semantic-segmentation-on-cityscapes)](https://paperswithcode.com/sota/real-time-semantic-segmentation-on-cityscapes?p=enet-a-deep-neural-network-architecture-for)

[![HitCount](http://hits.dwyl.io/soumik12345/Enet.svg)](http://hits.dwyl.io/soumik12345/Enet)

Pytorch Implementation of ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation (https://arxiv.org/abs/1606.02147). It is currently the 13th Best Model for [Real-Time Semantic Segmentation on Cityscapes test     ](https://paperswithcode.com/sota/real-time-semantic-segmentation-on-cityscapes).

![Training Results](./assets/enet_camvid_gif-1.gif)

### Training Notebooks

**Note:** Try using the Binder badge to inspect the notebooks.

- [Enet with PReLU Encoder on CamVid](https://github.com/soumik12345/Enet/blob/master/Enet_CamVid.ipynb)

- [Enet with Mish Encoder on CamVid](https://github.com/soumik12345/Enet/blob/master/Enet_CamVid_Mish.ipynb)

## Training Results on [CamVid Dataset](https://www.kaggle.com/jcoral02/camvid)

### Inference on Training Data

![Training Inference Results](./assets/image_train.png)

### Inference on Validation Data

![Validation Inference Results](./assets/image_val.png)

## TODO

- [x] Implement Vanilla Enet Architecture

- [x] Encorporate Custom Activations for Codebase

- [x] Train Enet on CamVid

- [x] Train Enet with Mish Encoder on CamVid

- [x] Experiment to find best Mish Version of Enet -> PReLU encoder + Mish Decoder

- [ ] Repeat Same experiments for Cityscapes Dataset

- [ ] Repeat Same experiments for SUN RGB-D Dataset

- [ ] Implement Lovasz Softmax Loss

## References

- [ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation](https://arxiv.org/abs/1606.02147)

- [Tensorflow Enet](https://github.com/kwotsin/TensorFlow-ENet)

- [ENet on TowardsDataScience](https://towardsdatascience.com/enet-a-deep-neural-architecture-for-real-time-semantic-segmentation-2baa59cf97e9)

- [Mish: Self Regularized Non-Monotonic Activation Function](https://github.com/digantamisra98/Mish)

- [Mish on Arxiv](https://arxiv.org/abs/1908.08681)