Flora Fix
Empowering People with AI & Ayurveda
The problem Flora Fix solves
i)Plant diseases cause up to 40% crop loss worldwide each year.
ii)Farmers & home growers often detect issues too late.
Lack of affordable, easy-to-use tools for early diagnosis.
Over-reliance on chemical pesticides harms soil & health.
iii)Limited awareness of Ayurvedic/natural remedies.
Challenges we ran into
🌱 Challenges I Ran Into
Dataset Collection & Quality
Finding a large, diverse, and high-quality dataset of diseased plant leaves.
Many datasets were unbalanced (some diseases had thousands of images, others very few).
Images varied in lighting, angle, and background, which made training harder.
Model Selection & Training
Tried multiple CNN architectures (VGG19, ResNet50, DenseNet, AlexNet, MobileNetV2).
Balancing accuracy vs. speed was difficult — heavy models gave high accuracy but slowed down prediction on web server.
Overfitting occurred due to limited dataset size for some classes.
Deployment with Django
Converting trained deep learning model (.keras / .h5) into a format compatible with Django backend.
Handling large model file size (loading time, server memory usage).
Ensuring TensorFlow/Keras dependencies worked smoothly on deployment environment.
Frontend–Backend Integration
Making sure image upload from HTML/JS frontend was properly passed to Django backend for model prediction.
Handling file path issues (local vs. server).
Displaying results dynamically without page reload.
Performance Optimization
Prediction speed was slow on CPU-only servers.
Had to compress and optimize model (chose VGG19 over MobileNetV2 for lightweight inference).
Accuracy vs. Real-world Use
Model performed well on dataset but sometimes misclassified real-world images taken by phone cameras.
Needed preprocessing (resizing, background cleaning) to improve real-world accuracy.
Explain ability of Predictions
Farmers/gardeners need to trust the diagnosis, but deep learning models are black boxes.
Hard to provide “why” a leaf was classified as diseased.
Natural Remedies Integration
Collecting reliable Ayurvedic/natural treatment data was challenging (limited scientific documentation).
Needed to simplify explanations so non-technical users could understand remedies.
Version Control & Collaboration
Managing GitHub repo with large ML files (models, datasets) was tricky.
UI/UX Challenges
Making a clean, farmer-friendly interface.