Osprey is a dynamic, multi-stream video analytics platform built on NVIDIA DeepStream 8.0. Add and remove RTSP/file sources at runtime via REST API — each stream gets GPU-accelerated inference (YOLO-det, YOLO-seg, or RT-DETR), multi-object tracking, and an independent RTSP output with zero downtime

The Purpose of this repository is to create a DeepStream/Triton-Server sample application that utilizes yolov7, yolov7-qat, yolov9 models to perform inference on video files or RTSP streams.

Sample face-recognition app using Deepstream 6.0.1 with FaceNet on Jetson Nano (L4T Jetpack 4.6.1)

Guide to set up DeepStream pyds and run an RTSP pipeline in Python. Note: NVIDIA Smart Recording isn’t in official Python bindings and is supported only in C++. This repo includes a compatible wheel and tweaks so you can install it easily and enable Smart Recording from Python.

This repository provides a custom implementation of parsing function to the Gst-nvinferserver plugin when use YOLOv7/YOLOv9 model served by Triton Server using the Efficient NMS plugin exported by ONNX.

Drop-in cv2.VideoCapture replacement with a selectable decode backend (CPU, CUDA/NVDEC, OpenCV, or DeepStream) for RTSP and file streams.

CoBaoHieu Master Computer Science's Thesis