Google Git
Sign in
chromium/chromium/src/d232a5d77ae4b065bc1905a4942a518aa5e07c4d/./remoting/test/frame_generator
tree: c7c9c8d11055f946021822a46652fe70170a5411
  1. scenarios/
  2. BUILD.gn
  3. differ_frame_generator.cc
  4. differ_frame_generator.h
  5. file_frame_generator.cc
  6. file_frame_generator.h
  7. frame_generator.h
  8. frame_generator.py
  9. headless_frame_generator.cc
  10. headless_frame_generator.h
  11. README.md
remoting/test/frame_generator/README.md

Frame Generator Tool

This directory contains a standalone tool for generating synthetic frame sequences using Headless Chrome. These sequences are used to stress-test and benchmark video encoders (like AV1) in the Chrome Remote Desktop context.

Requirements

Note: This tool is currently supported on Linux and Mac only.

Overview

The tool uses the Chrome DevTools Protocol (CDP) to drive a headless instance of Chrome. It loads a “scenario” (an HTML/JS file), advances time deterministically, and captures screenshots for each frame.

Usage

python3 remoting/test/frame_generator/frame_generator.py \
    --scenario <name> \
    --out-dir <path> \
    --frames 100 \
    --fps 30 \
    --width 1920 \
    --height 1080

Arguments

  • --scenario: The name of the HTML file in the scenarios/ directory (without .html).
  • --out-dir: Directory where the generated PNG frames will be saved.
  • --frames: Number of frames to generate (default: 100).
  • --fps: Frame rate for the generated sequence (default: 30.0).
  • --width/--height: Dimensions of the capture window (default: 1920x1080).
  • --chrome-path: Path to the Chrome binary (default: out/Release/chrome).

Adding New Scenarios

When creating a new scenario in scenarios/, you must adhere to the following technical contract:

  1. File Format: Must be a standalone HTML file.
  2. Deterministic Timing: You MUST implement window.setBenchmarkTime(t). This function is called by the Python script to advance the scene to a specific point in time t (in seconds).
  3. Seeded Randomness: Do NOT use Math.random(). Use a deterministic seeded random number generator (e.g., an LCG) to ensure identical output across different runs and machines.
  4. Visual Verification: You SHOULD implement an animate(t) loop using requestAnimationFrame that calls your render logic. This allows the scenario to “auto-play” when opened manually in a browser for debugging.
  5. No External Assets: Prefer procedurally generated content (Canvas/CSS) over external images to ensure the test is self-contained and loads instantly.

How it Works

  1. Orchestration: The Python script launches Chrome with --remote-debugging-pipe.
  2. Synchronization: For each frame, the script calls a global window.setBenchmarkTime(t) function within the loaded HTML page. This ensures that even animations are deterministic relative to the frame index.
  3. Capture: After the JS has updated the DOM/Canvas, the tool issues a Page.captureScreenshot command to get the raw pixel data.