Real-Time Video Upscaling Software -- 2

Замовник: AI | Опубліковано: 06.10.2025
Бюджет: 8 $

I need a piece of software that can take an incoming HLS stream and upscale it on-the-fly, running smoothly on Windows without noticeable added latency. The goal is to boost resolution and visual sharpness in real time so the upscaled output can be played, recorded, or forwarded just as the original stream would be. Key points • Target platform: Windows (desktop environment, recent 64-bit builds). • Input: HLS streaming video only; local files or other protocols are out-of-scope for this first release, though a clean architecture that could later plug in RTSP or DASH would be appreciated. • Performance: true real-time operation is essential—no buffering that drifts the stream. Leveraging GPU acceleration (CUDA, DirectX, Vulkan, OpenCL, or similar) is encouraged if it helps hit frame-rate. • Quality: the upscaled feed should be measurably sharper than basic bilinear/bicubic resizing. If you plan to use AI or deep-learning models (e.g., ESRGAN, Real-ESRGAN, TensorRT, ONNX Runtime) please outline which model, how it will be packaged, and its VRAM footprint. • UI/UX: a simple window is fine—just let me paste an HLS URL, pick an output resolution, and hit “Start”. Command-line control is acceptable as long as usage is clearly documented. Deliverables - Compiled Windows executable with any required DLLs - Full, buildable source code - Brief setup & usage guide (readme or PDF) Acceptance criteria 1. Upscaled stream plays continuously at the source frame-rate on a mid-range GPU (we’ll test on a GTX 1660). 2. Visual improvement over default FFmpeg scale filter is obvious on side-by-side playback. 3. Audio (if present in the HLS) remains in sync. If you have prior experience with FFmpeg filters, NVENC/AMD AMF, or custom CNN upscalers on Windows, please mention it when you bid—seeing demos or recorded comparisons will help me choose quickly.