ResNext-Reimplementation

This project is UTAR UCCD3074 Deep Learning For Data Science assignment. It is the reimplementation of ResNext-50 tested on pikachu dataset for binary image classification.

Pikachu dataset contains image that was web scraping from the internet using Azure Bing Web Search API. Visit Here (To Kaggle) for more information.

Result

compared against PyTorch official model

---

Notes

• The models directory is not able to upload to repository due to the file size is too big, use link: https://drive.google.com/drive/folders/13O0dE4QRiHCzvWlP2BX9ACmet5DjQZKV?usp=sharing

• Put the models folder in the project root directory

• Include only latest trained model is included to avoid the project file size going too big

• Use Application.ipynb to use the pretrained model and read the previous training statistics

• CIFAR10_ResNext_Deep_Learning.ipynb is the file for training CIFAR10 dataset

• Pikachu_ResNext_Deep_Learning.ipynb is the file for training Pikachu dataset

• models directory store all the pretrained model according to epoch, the .json file in the directory store the stats for previous training

• runs directory is for TensorBoard used

• pikachu dataset directory consists of pikachu binary classification image

• CIFAR10 test is incomplete and deprecated