pal.ai

pal.ai is a progressive web app that uses images of rice leaves to identify rice leaf diseases.

In the back end, we use a minimal SvelteKit server that proxies the Hugging Face Hosted Inference API, which interfaces with a model that uses a BEiT vision transformer model to classify images of rice plants for the presence/absence of diseases. The model is pre-trained using ImageNet 22k then finetuned using the PH Rice Leave Diseases dataset. The model achieves 95% accuracy using the dataset test set despite having only 1120 training data points.

• Kaggle Notebook: Classification by Fine-tuning BEiT
• Hugging Face Hub Link for the Model: jkrperson/Beit-for-rice-disease
• Weights & Biases Experiments: jkrperson/huggingface

In the front end, pal.ai is an installable progressive web app (PWA) written in TypeScript with the Svelte framework. It uses the Vite build system for asset bundling, packaging, and optimization. The PWA allows users to upload their own images of rice plants for classification.

Environment Variables

Project Members

• Macaraeg, Marc Gabriel C.
• Ortiz, Sebastian Luis S.
• Raborar, Angelica Julianne A.
• Resabal, Jose Kristian K.