Computer Vision & Object Detection Learning Roadmap

Welcome to the Object Detection Learning Roadmap repository. This project is structured as a bottom-up, hands-on journey from the most fundamental classic machine learning computer vision concepts to state-of-the-art transformer-based deep learning object detectors.

Each module in this repository is designed as a self-contained learning unit, focusing on a specific milestone in the evolution of object detection.

---

The Learning Path

graph TD
    classDef planned fill:#1e293b,stroke:#475569,stroke-width:2px,stroke-dasharray: 5 5,color:#94a3b8;
    classDef active fill:#1e3a8a,stroke:#3b82f6,stroke-width:3px,color:#eff6ff;
    classDef done fill:#064e3b,stroke:#10b981,stroke-width:3px,color:#ecfdf5;

    Step1["Step 1: R-CNN & Sliding Window Detectors (Done)"]:::done
    Step2["Step 2: Region Proposals (Fast R-CNN concept) (Current)"]:::active
    Step3["Step 3: Core Architecture (IoU + Anchors + NMS)"]:::planned
    Step4["Step 4: Faster R-CNN (Two-Stage Detectors)"]:::planned
    Step5["Step 5: YOLOv8 / YOLOv11 (One-Stage Real-Time Detectors)"]:::planned
    Step6["Step 6: Mini-YOLO (From Scratch)"]:::planned
    Step7["Step 7: Advanced Vision Transformers (DETR, RT-DETR)"]:::planned
    Step8["Step 8: Instance Segmentation & Multi-Object Tracking"]:::planned
    Step9["Step 9: Production Optimization (ONNX + TensorRT)"]:::planned
    Step10["Step 10: Advanced End-to-End Applications"]:::planned

    Step1 --> Step2
    Step2 --> Step3
    Step3 --> Step4
    Step4 --> Step5
    Step5 --> Step6
    Step6 --> Step7
    Step7 --> Step8
    Step8 --> Step9
    Step9 --> Step10

Loading

---

Directory Index & Roadmap Progress

Step	Module Name & Folder	Status	Key Focus Area	Preview	Live App
1	1-R-CNN+Sliding-Window(Detectors)	Complete	Classic brute-force sliding window search vs. region proposal (R-CNN) detectors, stride/window hyperparams, custom ResNet50 classifier model training on MNIST, batched PyTorch GPU/CPU inference, and Non-Maximum Suppression (NMS)		Yes (Flask + HTML5 Canvas)
2	2-Region-Proposals(Fast_R-CNN_Idea)	In Progress	Selective Search, structural bounding box proposal generation, once-per-image feature extraction, and RoI Pooling concepts		Planned
3	3-IoU+Anchors+NMS(Core-Concepts)	Planned	Mathematical IoU implementation, anchor box shape aspect ratios, and full vector-based multi-class NMS implementation from scratch		Planned
4	4-Faster-R-CNN	Planned	Fully two-stage detector pipelines, Region Proposal Networks (RPN), bounding box regression loss, and torchvision fasterrcnn training		Planned
5	5-YOLO(Single-stage-detection)	Planned	One-stage end-to-end grid prediction, objectness confidence scores, and training ultralytics models on custom datasets		Planned
6	6-Mini-YOLO(From-Scratch)	Planned	Custom single-stage neural network architecture, multi-part loss function (coordinate loss + class loss + objectness loss)		Planned
7	7-Advanced-Detection(Transformers)	Planned	Attention-based detectors, DEtection TRansformers (DETR), set prediction loss, bipartite matching, and anchor-free models		Planned
8	8-Tracking+Segmentation	Planned	Pixel-level semantic instance masks (Mask R-CNN), Kalman filtering, DeepSORT, and multi-object linear tracking pipelines		Planned
9	9-Production-Systems	Planned	Model conversions to ONNX format, TensorRT FP16/INT8 quantizations, dynamic batching, and high-performance C++/Python APIs		Planned
10	10-Advanced-Applications	Planned	Large-scale smart multi-camera surveillance system, edge deployments, and COCO competition dataset fine-tuning		Planned

---

Getting Started with Step 1: R-CNN & Sliding Window Detectors

The first folder contains a complete, interactive visualization tool showing how standard image classifiers are adapted into object detectors using both brute-force sliding window search and selective search region proposals (the core concept of R-CNN).

Quick Start:

1. Ensure you have Miniconda/Anaconda installed.
2. Open your terminal and enter the first step's folder:

   cd "1-R-CNN+Sliding-Window(Detectors)"

3. Create the conda environment and activate it:

   conda create -n sliding_window_env python=3.9 -y

4. Run the automated startup script:

   chmod +x run.sh
   ./run.sh

5. Open your web browser and navigate to: http://127.0.0.1:5001

Read the full project detail and theory inside the Step 1 README.

---

License

This repository is licensed under the MIT License.