This repository illustrates the steps for training YOLOv3 and YOLOv3-tiny to detect fire in images and videos. First, a fire dataset of labeled images is collected from the internet. The images with their annotations have been prepared and converted into YOLO format and put into one folder to gather all the data. Therefore, the data folder contains images ('*jpg') and their associated annotations files ('.txt') with the same name.

Conversion utility for Label Studio video annotations to a YOLO-compatible format, including bounding box interpolation

Converts Label Studio video annotations to YOLO format, solving the issue of missing native support for YOLO exports, and making it easier to use your annotations for YOLO-based training.

Extracts frames from videos, allows an interface to draw bounding boxes and it outputs a text file with annotations for YOLO dataset

Processes video footage to automatically detect, encode, and cluster faces without requiring labeled datasets. Using OpenCV, dlib, and DBSCAN, it extracts facial embeddings, groups faces by identity, and generates organized datasets with YOLO-style annotations.

This project implements a real-time object detection system using a custom YOLO model to track players, referees, and the ball in recorded soccer footage. It generates live annotations, including ball possession, player speed, and distance traveled, providing key insights directly on the video feed.

A Streamlit application that uses a trained YOLO model to detect vehicles in uploaded videos, count the detected classes, and generate a processed video with annotations.

This project implements a YOLO v5 model using PyTorch for graffiti detection. It converts annotations to YOLO format, trains iteratively, and achieves accurate detection in images and real-time video

This notebook delivers a full end-to-end pipeline using Autodistill + Grounded SAM to generate segmentation-based annotations from video frames, convert them to YOLO bounding boxes, and fine-tune a YOLOv8 model all inside a single environment.

This project implements player detection, tracking, and re-identification in sports videos using a YOLO-based object detector and ByteTrack. It assigns unique IDs to players and maintains consistent identities, even when players leave and re-enter the frame, providing clear visual annotations for players, referees, and the ball.

Vehicle Registration Plate Detection using YOLOv8m trained on 5,308 images. Converted Pascal VOC annotations to YOLO format, validated ground truth visually, and evaluated using COCO metrics. Achieved ~0.92 mAP@50 with strong precision and recall. Includes training pipeline, evaluation, and real-time image/video inference for ANPR applications.