This repository focuses on re-implementing classic and influential Deep Learning research papers, mainly in the field of Computer Vision, with the goal of deeply understanding their core ideas and training details rather than just using existing libraries.
- Re-implementation of well-known DL architectures from original papers
- Emphasis on clarity, correctness, and educational value
- Experiments are primarily conducted on CIFAR-10
- ResNet
- DenseNet
(More architectures will be added over time.)
- CIFAR-10
The dataset is chosen for fast experimentation and fair comparison between different architectures.
-
Add more paper implementations in computer vision
-
Write accompanying blog posts explaining:
- Theoretical background
- Architectural design choices
- Training tricks and pitfalls
- Personal insights from re-implementation
The purpose of this project is to strengthen understanding of deep learning models by building them from scratch and bridging the gap between research papers and practical implementation.