A curated collection of end-to-end Machine Learning & Deep Learning projects spanning Computer Vision, Audio Classification, and more.
This repository showcases hands-on ML/DL projects built from scratch — covering real-world problem statements, custom-trained models, and practical implementations. Each project demonstrates skills in data preprocessing, model architecture design, training pipelines, and evaluation.
| # | Project | Domain | Key Techniques |
|---|---|---|---|
| 1 | 🐦 Audio Classifier — Capuchin Calls | Audio / Signal Processing | CNN, Spectrograms, Binary Classification |
| 2 | 🔥 Class Activation Map | Explainable AI | GradCAM, Feature Visualization |
| 3 | ✨ Image Super Resolution | Generative Models | U-Net, Upsampling |
| 4 | 🎭 Image Segmentation | Computer Vision | U-Net, Semantic Segmentation |
| 5 | 📦 Object Detection | Computer Vision | YOLO, Bounding Boxes, mAP |
Detecting Capuchin bird calls from raw audio using deep learning.
- Converts audio waveforms into mel spectrograms for visual-based classification
- Binary classifier trained to distinguish Capuchin calls from background noise
- Built with TensorFlow / Keras using a CNN architecture
- Evaluated on real-world forest recordings
Tech Stack: Python TensorFlow Librosa NumPy Matplotlib
Making deep learning models interpretable using visual explanations.
- Implements Grad-CAM to highlight regions in an image that influenced the model's prediction
- Applied to pre-trained CNNs (ResNet, VGG) on classification tasks
- Helps in debugging model behavior and understanding what the model "sees"
- Useful for model auditing and trust in AI systems
Tech Stack: Python PyTorch / TensorFlow OpenCV Matplotlib
Enhancing low-resolution images to high-resolution outputs using deep learning.
- Implemented U-Net architecture with skip connections for accurate detail reconstruction
- Encoder-decoder design preserves spatial information lost during downsampling
- Benchmarked using PSNR and SSIM metrics
- Demonstrates how U-Net's dense skip connections recover fine-grained textures and edges
Tech Stack: Python TensorFlow / PyTorch OpenCV Pillow
Pixel-level scene understanding for precise region identification.
- Implements U-Net architecture for semantic segmentation
- Trained to classify every pixel in an image into meaningful categories
- Evaluated with IoU (Intersection over Union) and Dice coefficient
- Applicable in medical imaging, autonomous driving, and satellite imagery
Tech Stack: Python TensorFlow / Keras OpenCV NumPy
Locating and classifying multiple objects within images in real time.
- Built using YOLO (You Only Look Once) for fast, accurate detection
- Trained and fine-tuned on custom datasets with bounding box annotations
- Evaluated using mAP (mean Average Precision)
- Supports real-time inference on images and video streams
Tech Stack: Python YOLOv5 / YOLOv8 OpenCV PyTorch
⭐ If you find this repository useful, please consider giving it a star! ⭐
Made with ❤️ and lots of GPU hours