Yolov1-from-scratch

A concise, from-scratch implementation of YOLOv1 for object detection using PyTorch. This repository contains a minimal dataset, training loop, loss, and utilities to experiment and visualize results.

---

Features

• Educational implementation of YOLOv1 architecture (Yolov1 in model.py)
• Custom YOLO loss implemented in loss.py
• Lightweight dataset loader and utilities to compute mAP, NMS, and visualize predictions (data.py, utils.py)
• Example training and evaluation script: train.py

---

Requirements

• Python 3.8+
• PyTorch
• torchvision
• numpy, pillow
• tqdm

Install with pip:

pip install torch torchvision numpy pillow tqdm

---

Quick start

1. Check the images and labels in archive/images and archive/labels (VOC-style text labels).
2. Edit hyperparameters in train.py (e.g., BATCH_SIZE, EPOCHS, LEARNING_RATE).
3. Run training:

python train.py

The script will train the model and print the mAP after each epoch; a checkpoint will be saved if performance is high.

---

Visualizations

Architecture & Loss

YOLOv1 Architecture	YOLOv1 Loss Function

Prediction Examples

Example 1	Example 2

Note: Replace LOAD_MODEL_FILE in train.py to visualize saved checkpoints or set LOAD_MODEL = False to train from scratch.

---

Notes

• This project targets clarity and learning rather than production performance.
• Feel free to extend the dataset, add augmentations, and tune the loss/architecture.

---