This project is part of HELLO FOSS 2022 by WnCC.
We have build a Deep learning model to perform an image classification task
The pre - equisites for you to understand and contribute in this project are Machine learning, understanding of Neural networks, Python and PyTorch library.
Don't worry, if you dont know the above tech stack, you can learn them with the resources we'll link down and still contribute!
We have used ResNet9 architecture for our model, about which you can read more here.
We have imported the dataset from Kaggle, it's divided into 4 classes having almost 5k images and also it's fairly balanced as almost equal number of images for all the 4 classes of batting shots.
Try exploring the dataset on your own to gain insights which eventually help you to tune the model to achieve higher accuracy.
Good First Issue is an initiative to artifically create easy to fix issues in projects, so developers who've never contributed to open-source can get started quickly.
- Try changing number of epochs, you might be able to see accuracy isn't saturated yet.
- Correct the values of normalisation transformation. Calculate mean and standard - deviation for each channel and fill in the correct values.
- It's always better to visualize results graphically, plot graps for various paramters like loss vs no. of epochs, related to accuracy etc.
- Try finding the ideal batch-loader size by changing it's value and running the model.
https://drive.google.com/drive/folders/1mTojuq6HrYSX01Dif6-3WePFKAflkaoQ?usp=sharing