A collection of computer-vision object-detection projects exploring a range of frameworks and detector architectures — from classic Haar cascades to modern single- and two-stage deep learning detectors (YOLOv4-tiny, YOLOv5, YOLOv8, and Detectron2's Faster R-CNN). Each project lives in its own directory with its own README, dataset notes, and runnable notebook/script.
| Project | Task | Framework / Model |
|---|---|---|
| Blood Cell Detection | Detect red/white blood cells and platelets in microscope images (BCCD dataset) | YOLOv5 (Ultralytics) |
| Mushroom Type Detection | Detect and classify mushrooms (North American Mushrooms dataset) | Detectron2 — Faster R-CNN (X-101 FPN) |
| Pothole Detection | Detect potholes in road images for road-condition monitoring | Tiny YOLOv4 (Darknet) |
| Car Detection | Detect cars from CSV-annotated images, with full annotation-conversion pipeline | YOLOv8 (Ultralytics) |
| License Plate Detection & Blurring | Detect and anonymize vehicle license plates | OpenCV Haar cascade |
Folder names above are placeholders — rename the links to match your actual directory names.
Across these projects the repo touches the main building blocks of an object-detection workflow:
- Multiple detector families — one-stage (YOLOv4-tiny, YOLOv5, YOLOv8), two-stage (Faster R-CNN), and a classical cascade classifier.
- Dataset preparation — converting between annotation formats (CSV → YOLO, COCO JSON, YOLO Darknet), normalizing bounding boxes, and building train/validation/test splits.
- Training & transfer learning — fine-tuning from pretrained backbones and training custom configs driven by class count.
- Evaluation — mAP / COCO metrics, loss and accuracy curves, and qualitative inference on held-out images.
- Tooling — PyTorch, Ultralytics, Detectron2, Darknet, OpenCV, and Roboflow for dataset management.
Object-Detection-Projects/
├── Blood-Cell-Detection-YOLOv5/
├── Mushroom-Detection-Detectron2/
├── Pothole-Detection-TinyYOLOv4/
├── Car-Detection-YOLOv8/
├── License-Plate-Detection/
└── README.md
Each subdirectory contains its own README.md with dataset, model, requirements, and usage details specific to that project.
- Clone the repository:
git clone https://github.com/pouyaardehkhani/Object-Detection-Projects.git cd Object-Detection-Projects - Open the project you're interested in and follow its README.
- Most deep-learning projects are notebook-based and assume a CUDA-enabled GPU (Google Colab works well).
- Several of the deep-learning projects are adapted from open tutorials (e.g. Roboflow's custom-training guides) and applied to different datasets and use cases.
- Framework versions differ between projects (for example, the Detectron2 project pins older CUDA/torch versions); see each project's README for environment specifics.