Yolov1_from_scratch

This project is an implementation of the YOLO (You Only Look Once) object detection algorithm using PyTorch. The purpose of this project is to provide a basic understanding of how YOLO works and how to implement it from scratch. The code has been done by following the [2] video.

Dependencies

matplotlib==3.7.1
numpy==1.24.2
pandas==1.5.3
Pillow==9.4.0
torch==1.13.1
torchvision==0.14.1
tqdm==4.65.0

Files

config.py: A configuration file containing hyperparameters for the model.
model.py: A Python module containing the YOLO implementation.
train.py: A Python script for training the YOLO model.
loss.py: A Python script for calculating the loss
dataset.py: A Python script for accessing the datas

Dataset

The dataset used is the PascalVoc YOLO and is available on Kaggle here The different models have been trained on the "8examples.csv" and "100examples.csv" for personal learning purposess.

Usage

Training the model

To train the YOLO model, run the following command:

python train.py --dir /path/to/data

dir: The path to the directory containing the training images and annotations.

Training the model on Colabs

Follow the instructions in the ‘train_on_colabs.ipynb' notebook.

Loss function

Architecture

Results

The model has only been trained on the training data and is therefore overfitting. The goal of this project is to understand how the model works, not to train a real model from scratch.

Notes:

• this has been done by following the excellent tutorials from Aladdin Persson [2], the original paper [1] and tanjeffreyz's repository[3].
• the generated models are not maid to be used as they clearly overfit the training data.

References

[1] - YOLOv1 Paper

[2] - Aladdin Persson's video that has been followed to create this project

[3] - tanjeffreyz's repository