PyTorch-CIFAR-10-autoencoder

KO | EN

gitlite — search

#python #php #dotfiles #arduino #linux #android #javascript #nodejs #docker #atom #ios #bot

PyTorch-CIFAR-10-autoencoder

★ 75 Open GitHub ↗

This is a reimplementation of the blog post "Building Autoencoders in Keras". Instead of using MNIST, this project uses CIFAR10.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# building-autoencoders-in-Pytorch
This is a reimplementation of the blog post "Building Autoencoders in Keras". Instead of using MNIST, this project uses CIFAR10.

## Current Results (Trained on Tesla K80 using Google Colab)
First attempt: (BCEloss=~0.57)  
![decode](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/colab_predictions.png)

Best Predictions so far: (BCEloss=~0.555)  
![decode](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/colab_predictions2.png)
![decode](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/colab_predictions22.png)

Targets:  
![target](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/colab_tar.png)
![target](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/target2.png)

## Previous Results (Trained on GTX1070)
First attempt: (Too much MaxPooling and UpSampling)  
![decode](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/decoded_img.png)

Second attempt: (Model architecture is not efficient)  
![decode](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/decoded_img2.png)

Targets:  
![decode](https://raw.githubusercontent.com/chenjie/PyTorch-CIFAR-10-autoencoder/master/weights/target.png)

## License
[MIT](LICENSE)

PyTorch-CIFAR-10-autoencoder

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README