This project is one of the miniprojects of our Deep Learning EE-559 Course at EPFL. Our GitHub Repo can be found
here.
Overview
I implemented a PyTorch Deep Learning framework (including the backprop of modules without using autograd). Implemented Modules are:
- Conv2d
- Upscale Module (by Nearest Neighbor Sampling + Convolution)
- Nonlinearities: ReLU, Sigmoid
- Optimizer: SGD Optimizer
- Sequential Module
- MSELoss Module
- Weight saving/loading modules
- Device selecting module (train/test on CPU/GPU)
Furthermore, I built two models resemble Noise2Noise for the task of image denoising.
Results
The final results achieved by 
