Research-Papers-Implementation

This repository contains the paper implementations of various advanced CNN based architectures. The primary goal of this repository is to show how to implement these papers using tensorflow-keras. Model performance and accuracy are not taken into consideration while implementing these papers. I just focused on the implementation part.

For the implementation, I used the facial emotion dataset. The reason for not opting for other well-known datasets like MNIST is due to the fact that those datasets are fairly simple. Whereas facial emotion recognition aka FER is a bit complex task to do.

Currently, I have implemented the below papers, for each paper I mentioned the link to the paper as well. Also, I didn't exactly mimic the architecture because they are very deep and are usually designed for images with larger sizes like in imagenet but the dataset I used is of size 48x48 which is around 5 times smaller then imagenet. Hence instead of using the same number of layers as mentioned in the paper I just decreased the number of layers. But the model architecture is exactly the same and if you wish you can play by adding more and more layers, it's really easy to do in my code.

I have implemented each paper in separate google-colab (to leverage free GPU). For each paper, I mentioned the link to the paper, notebook in this repository, and colab notebook as well. You can directly read the notebooks in this repository but if you face problems while rendering jupyter notebooks in GitHub(which you may) then directly open the colab link.

Papers Implemented

1. DenseNets
   paper - https://arxiv.org/abs/1608.06993v3
   notebook - https://github.com/greatsharma/Research-Papers-Implementation/blob/master/DenseNet.ipynb
   colab - https://drive.google.com/file/d/1qjUdDKxAIdpMHc9by81XTWp9ZNzyuJQc/view?usp=sharing
   

2. GoogLeNet / InceptionNet
   paper - https://arxiv.org/abs/1409.4842
   notebook - https://github.com/greatsharma/Research-Papers-Implementation/blob/master/InceptionNet.ipynb
   colab - https://drive.google.com/file/d/1FNUTbp9735lRMWQYempPEhgbhfmEBbCP/view?usp=sharing
   

3. ResNet
   paper - https://arxiv.org/abs/1512.03385
   notebook - https://github.com/greatsharma/Research-Papers-Implementation/blob/master/ResNet.ipynb
   colab - https://drive.google.com/file/d/1NW42aw5f83TA-AyADs-yY1BdzV35PAcK/view?usp=sharing
   

4. XceptionNet
   paper - https://ieeexplore.ieee.org/document/8099678
   notebook - https://github.com/greatsharma/Research-Papers-Implementation/blob/master/XceptionNet.ipynb
   colab - https://drive.google.com/file/d/1Bo92DVpz1KyYPTCdC3neGbj8wJFdks0o/view?usp=sharing
   

5. U-Net
   paper - https://arxiv.org/abs/1505.04597
   notebook - https://github.com/greatsharma/Research-Papers-Implementation/blob/master/denoising_Unet.ipynb
   colab - https://colab.research.google.com/drive/1ubWe7wj_K7_qGDostLnzP7Wf4DYqhgN6?usp=sharing
   

If you face any problem while rendering notebook in github or opening in colab then please try it 3-4 times and if the situation remains the same then simply open an issue, I will fix that.