AetherVision: Image Based Weather Classification 🌦️
AetherVision is a high-performance computer vision framework designed to classify meteorological conditions with precision. Leveraging Vision Transformers (ViT) and custom Focal Loss optimization, the system achieves state-of-the-art accuracy in handling imbalanced environmental datasets.
🚀 Key Features
State-of-the-Art Architecture: Powered by timm (PyTorch Image Models) using pre-trained Vision Transformers for global contextual awareness.
Imbalance Mitigation: Implements custom Focal Loss to ensure robust performance on minority weather classes.
Interpretability: Built-in Grad-CAM integration to visualize model attention and debug feature identification.
Production-Ready Pipeline: Optimized data loaders with seamless support for large-scale image processing and hardware acceleration (CUDA).
📊 Results & Visualization
The AetherVision system has been optimized to ensure high-fidelity classification and reliable feature identification. Below is the performance trajectory of the model and a sample visualization of its decision-making process.
Performance Metrics: The graph above illustrates the model's convergence, showing minimal divergence between training and validation loss, confirming the system's ability to generalize to unseen data.
Interpretability: Using Grad-CAM, we can visualize the model's attention. The heatmap below highlights how the Vision Transformer (ViT) effectively isolates meteorological features, ignoring extraneous background noise.
📋 Project Structure
🏗️ Technical Stack
This stack highlights your proficiency in deep learning, data engineering, and model interpretability.
🚀 Getting Started
This project is optimized for high-performance computing. Because training Vision Transformers (ViT) on local hardware can be time-prohibitive, this workflow is designed to run seamlessly on Google Colab with a T4 GPU.
Option A: Running on Google Colab (Recommended)
This workflow is optimized for the T4 GPU to significantly reduce training times.
Upload to Drive: Upload the entire AetherVision/ folder to your Google Drive.
Open Notebook: Open AetherVision.ipynb in Google Colab.
Configure Runtime: Navigate to Runtime > Change runtime type and select T4 GPU.
Mount & Setup: In the notebook, run the following cells:
from google.colab import drive drive.mount('/content/drive')
%cd /content/drive/MyDrive/AetherVision/
pip install -r requirements.txt -q
Execute Pipeline: python main.py
Option B: Running on Local Machine (More execution time)
If you prefer local environment:
- Clone the Repository:
git clone https://github.com/Anisca-hub/AetherVision.git
cd AetherVision
- Install Dependencies:
pip install -r requirements.txt
- Run the Pipeline:
Execute the main entry point to start training or evaluation:
python main.py
🎓 About the Author Built by [Anisca Jha] | [https://www.linkedin.com/in/anisca-jha-93ba83308]
Designed to showcase excellence in modern AI development, AetherVision implements high-performance Vision Transformers (ViT) and sophisticated interpretability modules (Grad-CAM). The project serves as a blueprint for robust machine learning workflows, emphasizing modular data engineering, hardware-accelerated training, and verifiable, reproducible model performance.