Upwork home
Sign up
Project Catalog
DesignDesign
Video & AudioVideo & Audio
Development & ITDevelopment & IT
MarketingMarketing
Writing & TranslationWriting & Translation
MoreMore

You will get CUDA NLM Video Denoiser Plugin for DeepStream 7.1 on Jetson Orin

Oleksii S.Status: Offline
Oleksii S. Oleksii S.
Preview image

Let a pro handle the details

Buy Other AI & Machine Learning services from Oleksii, priced and ready to go.
Oleksii S.Status: Offline
Oleksii S. Oleksii S.
Preview image

Let a pro handle the details

Buy Other AI & Machine Learning services from Oleksii, priced and ready to go.

Project details

You will get a production-ready GStreamer plugin for NVIDIA DeepStream 7.1 that removes video noise on Jetson Orin at 160 FPS (1080p). Unlike simple blurs or temporal averaging, my algorithm uses Non-Local Means (NLM) with Lucas-Kanade Optical Flow for motion compensation. This preserves fine textures and sharp edges while eliminating noise — no ghosting on moving objects.

The plugin operates in-place on NVMM memory (NvBufSurface) with zero CPU-GPU copies. All kernels run on CUDA, optimized for sm_87 (Jetson Orin AGX/NX/Nano).

Deliverable includes: compiled .so plugin, full source code, commented CUDA kernels, Makefile, and example pipelines. Tune parameters (h, patch/search radii) for your specific noise profile. Works with nvinfer, RTSP, USB cameras, and files.

Requirements: JetPack 6.x, CUDA 12.x, DeepStream 7.1.
AI Development Type
Deep Learning, Model Tuning, Software Maintenance
AI Tools
NVIDIA AI Platform, OpenCV
AI Development Language
C++
What's included
Service Tiers Starter
$500
Standard
$1,200
Advanced
$2,500
Delivery Time 5 days 10 days 15 days
Number of Revisions
123
AI Model Integration
-
-
-
Detailed Code Comments
-
Knowledge Graph
-
-
-
Model Documentation
-
-
Ontology
-
-
-
Source Code
Taxonomy
-
-
-

Frequently asked questions

Oleksii S.Status: Offline

About Oleksii

Oleksii S.Status: Offline
Edge AI Engineer | DeepStream | TensorRT | YOLO | Jetson
Kyiv, Ukraine - 12:37 am local time
Computer Vision / Machine Learning Engineer with a strong mathematical background and hands-on experience building and optimizing real-time computer vision systems on NVIDIA Jetson devices. Focused on object detection and image classification using YOLO, MobileNet, DeepStream, and TensorRT for efficient edge deployment.

Built 30 FPS real-time CV pipelines on NVIDIA Jetson Orin at <5W GPU power
Deployed YOLO + DeepStream + TensorRT systems for license plate OCR & people counting
Parallel inference optimization via Ray actors — 2–3× throughput improvement

Experienced in designing multi-camera video pipelines, accelerating neural network inference under resource constraints, and developing custom CUDA-based video processing filters. Implemented HDR video logic including multi-exposure handling, fusion, and tone mapping to improve image quality in real-time systems.

Additionally, I have a background in data analysis and statistical modeling (Python, SQL, SAS), enabling data-driven engineering decisions. I value performance, reliability, and practical deployment of machine learning models in production environments.

Steps for completing your project

After purchasing the project, send requirements so Oleksii can start the project.

Delivery time starts when Oleksii receives requirements from you.

Oleksii works on your project following the steps below.

Revisions may occur after the delivery date.

1

Client sends video sample, resolution, FPS, and Jetson Orin model (AGX/NX/Nano).

2

I set up DeepStream 7.1 environment and compile the plugin for target GPU.

Review the work, release payment, and leave feedback to Oleksii.