← Back to Hub

Project "Giant's Eye" - Plan

Web-Based Neural Super Resolution Architecture Report

📹📹

1. Feasibility & Technology Stack

This section evaluates the feasibility of executing DLSS-style neural rendering inside a web browser using Slang shaders on a video stream. It highlights the constraints of the web platform versus native hardware and defines the viable technological path.

The Reality of Web "DLSS"

True NVIDIA DLSS relies on proprietary Tensor Cores and closed-source drivers, which are inaccessible from a Web View. However, Neural Rendering and Spatial/Temporal Super Resolution (like FSR or custom CNNs) are highly feasible using WebGPU.

  • Slang Shader Language: Excellent choice. Slang can compile down to WGSL (WebGPU Shading Language) via WebAssembly, allowing you to write high-performance compute shaders for the web.
  • WebGPU Compute: Required. Standard WebGL cannot efficiently execute the matrix multiplications needed for neural upscaling.
  • Video Texture Mapping: HTML5 <video> elements can be zero-copy mapped directly to WebGPU textures using importExternalTexture.

Compute Capability Comparison

Estimated TFLOPS availability for ML/Rendering tasks.

2. The "Giant's Eye" Stereo Setup

This section analyzes your specific hardware setup: two stationary cameras with a massive 3-meter horizontal baseline, aimed at a 3-meter distance. Understanding this geometry is critical for utilizing the dual streams for enhancement.

The 3-Meter Baseline

Average human interpupillary distance is ~0.065m. A 3m baseline creates extreme hyperstereoscopy. The world will appear as a miniature model (the "Giant's Eye" effect).

Focal Point Convergence

Camera 1 at X=0. Camera 2 at X=3m. Both looking at a spot X=3m, Z=3m (assuming Z is depth towards field). This means Camera 1 is angled 45 degrees, while Camera 2 is looking straight ahead (90 degrees). *Assuming 50 yard line setup.

Super Resolution Advantage

The massive parallax provides incredibly distinct sub-pixel information for the shared frustum. By running optical flow/stereo matching across the two streams, your Slang compute shader can merge details from both views to reconstruct a higher-resolution master frame.

Top-Down View

3. WebGPU + Slang Pipeline Architecture

This interactive diagram details the flow of data from the remote video streams into the browser, through the Slang-compiled WebGPU compute shaders, and to the final display. Click the components to view technical details.

Slang Compute Pipeline

Select a pipeline step

Click on the flowchart blocks on the left to reveal the specific technical architecture and exact steps required for that stage of the web application.

// Awaiting selection...

5. Lab & Technology Tests

Active experimentation modules to validate individual components of the neural rendering pipeline.

📡

HLS Ingest & Bitrate Test

Validate multi-bitrate stream decoding and network stability up to 80Mbps.

🧠

WebGPU Shader Benchmark

Upcoming: Test CNN inference latency using Slang-compiled WGSL.