A capstone project as part of IBM artificial intelligence and cybersecurity course. Project domain is in artificial intelligence, specifically image classification. The purpose of the project is to classify images of five different fish species found in the Indian subcontinent as accurately as possible. The code and project is based on IBM Artificial Intelligence Patterns, specifically on "Build and image classification model" found on https://developer.ibm.com/patterns/build-an-american-sign-language-alphabet-classifier-using-pytorch-and-gpu-environments-on-watson-studio/
- Watson Studio: Used for code writing and model training, using 40 vCPU + 172 GB RAM + 1 NVIDIA V100 (1 GPU)
- Cloud Storage: Used for the storage of the dataset used for model training
The dataset used is acquired from Kaggle and is provided by Ritik Bompilwar. A total of 1033 images from 5 class is available for training, validating, and testing in a 70:20:10 ratio respectively.
Dataset link: https://www.kaggle.com/datasets/ritikbompilwar/fishes-species-in-the-indian-subcontinent
The code assumes that dataset folder structuring is as how it is structured in this repository. For easier usage, a portable version of the code is also provided. The portable code will download the dataset from Kaggle and restructure it to match the required structuring, all within the code without need of external management.