https://github.com/blankeos/cgjj-detector
🕵️♀️ The CGJJ or the Carlo Glecy Jessa Jonah Detector is an object detection model trained on YoloV5 to detect the faces of Carlo, Glecy, Jessa, and Jonah.
https://github.com/blankeos/cgjj-detector
Last synced: 2 days ago
JSON representation
🕵️♀️ The CGJJ or the Carlo Glecy Jessa Jonah Detector is an object detection model trained on YoloV5 to detect the faces of Carlo, Glecy, Jessa, and Jonah.
- Host: GitHub
- URL: https://github.com/blankeos/cgjj-detector
- Owner: Blankeos
- Created: 2022-11-29T03:00:52.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2022-12-05T05:44:18.000Z (over 3 years ago)
- Last Synced: 2025-04-07T03:54:52.547Z (about 1 year ago)
- Language: Python
- Size: 41.7 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 🕵️♀️ CGJJ Detector
### 🤔 About
The **CGJJ** or the **Carlo Glecy Jessa Jonah Detector** is an object detection model trained on YoloV5 to detect the faces of Carlo, Glecy, Jessa, and Jonah.
We annotated a custom 30FPS video dataset on [makesense.ai](https://makesense.ai/) for an object detection task of 4 classes: `carlo`, `glecy`, `jessa`, `jj`. We have 665 images for training and 100 images for testing.
This repository has two purposes:
1. [Use the model for inference](#🚀-get-started)
2. [Recreate the model from scratch](#😎-how-to-recreate-this-app-from-scratch)
Submitted to **Mr. John Christopher Mateo** for **CCS 250 Computer Vision** as a Final Project for the semester.
### 🚀 How to Use:
1. Clone this repository
```sh
$ git clone https://www.github.com/blankeos/cgjj-detector
$ cd cgjj-detector
```
2. Install the requirements
```sh
$ python -m venv venv # Optional: Create a virtual environment
$ pip install -r requirements.txt
```
3. Run scripts to use model for inference (You have 3 choices):
- [x] Create video from `./videoData/test.mp4` (Smoothest but Long Process Time)
```sh
python create_vid.py
# Saves video on 'output_video.mp4'
```
- [x] Use model in real-time on **camera** (Choppy depending on GPU)
```sh
python realtime_cam.py
```
- [x] Use model in real-time on `./videoData/test.mp4` (Choppy depending on GPU)
```sh
python realtime_vid.py
```
### 📝 Important Links
- [Final Annotated Dataset](https://drive.google.com/drive/folders/18rj0MZ2vT_22wnRXBmll7RkDSqnGJXXE?usp=share_link)
- [YoloV5 Training Notebook](https://colab.research.google.com/drive/1qct0OJryhOWcNPBE15wgIgECriVE5X75?usp=sharing)
- [Final Model Used](/model/cgjj_best.pt)
### 📁 Dataset Directory Structure
```
| - dataset
| -- images
| -- train
| -- val
| -- labels
| -- train
| -- val
labels.txt
```
### 🧠 Final Model Performance
```
Model summary: 157 layers, 7020913 parameters, 0 gradients, 15.8 GFLOPs
Class Images Instances P R mAP50 mAP50-95: 100% 4/4 [00:01<00:00, 2.30it/s]
all 100 400 0.996 0.997 0.995 0.775
carlo 100 100 1 0.99 0.995 0.789
glecy 100 100 0.998 1 0.995 0.689
jessa 100 100 0.999 1 0.995 0.815
jj 100 100 0.988 1 0.995 0.805
```
---
### 😎 How to recreate this project from scratch
Make sure you already cloned this repository and installed the requirements.
1. Run preprocessing script
```sh
python preprocessing.py
# This will use `./videoData` and create images following the YoloV5 directory structure
# Then, outputs the preprocessed data on `/cgjj-dataset`
```
2. Annotate the dataset on [makesense.ai](https://makesense.ai/) for an object detection task of 4 classes: `carlo`, `glecy`, `jessa`, `jj`. _(It took us 6 hours to do this)_ 😅
3. Upload the dataset on your Google Drive (To be trained on Colab).
4. Train the dataset on our **YoloV5 Training Notebook**.
5. Download the model from `runs/expt/best_weights.pt` in Colab
6. Put the `best_weights.pt` inside `/model` dir of this repository. Rename as `cgjj_best.pt`.
7. Go back to Step 3 in [Get Started](#🚀-get-started).