Few-shot Object Counting and Detection (ECCV 2022)

Uses eufy camera with rf-detr person detection to count swim laps

基于DETR-R101的多类别车辆检测与计数系统研究

No description available

Real-time animal detection and counting system using YOLO and DETR with a custom dataset.

Compares YOLOv12 (fast CNN) & RF-DETR (accurate Transformer) for real-time crowd counting. Evaluates accuracy (mAP) vs. speed (FPS) in varied densities & night scenes. YOLOv12 excels in speed for real-time use; RF-DETR in accuracy for dense crowds. Key trade-offs for crowd monitoring applications.

Compares YOLOv12 (fast CNN) & RF-DETR (accurate Transformer) for real-time crowd counting. Evaluates accuracy (mAP) vs. speed (FPS) in varied densities & night scenes. YOLOv12 excels in speed for real-time use; RF-DETR in accuracy for dense crowds. Key trade-offs for crowd monitoring applications

Compares YOLOv12 (fast CNN) & RF-DETR (accurate Transformer) for real-time crowd counting. Evaluates accuracy (mAP) vs. speed (FPS) in varied densities & night scenes. YOLOv12 excels in speed for real-time use; RF-DETR in accuracy for dense crowds. Key trade-offs for crowd monitoring applications

A comprehensive vehicle detection and tracking system implementing both DETR and Faster R-CNN approaches with DeepSORT for object tracking. Features real-time vehicle counting, traffic analysis, and visualization capabilities.

Engineered a real-time traffic surveillance system for vehicle detection, tracking, and counting. • Integrated RT-DETR for accurate and efficient vehicle detection across multiple frames. • Enabled geolocation visualization by linking detected data with Google Maps API.

Compact implementation of DETR (Tiny-DETR) for pedestrian detection on the PennFudanPed dataset. The project explores transformer depth, backbone choice, query count, and data augmentation under a small-dataset regime, achieving 0.51 mAP@0.5 with an end-to-end Hungarian matching–based detector.

An AI-powered system detects triple riding on two-wheelers using DETR for object detection, CNN for rider counting, and OCR for license plate recognition. It ensures real-time, scalable traffic monitoring with temporal validation to reduce false positives, enhancing enforcement, safety, and public accountability.

A computer-vision web app that identifies food items in a photo, estimates portion counts and total calories, stores your daily intake, and lets you download a PDF report. Built with YOLO-DETR, Streamlit and Python.