Filter Guide: Adaptive DeBlocker - Crowsinc/LiveVisionKit GitHub Wiki

A denoising filter purpose built to lessen light H.264 blocking artifacts in IRL streams.

An H.264 encoder operates on a grid of 16x16 pixels called macroblocks. When the bitrate is insufficient, their boundaries become visible and a distracting grid pattern (blocking artifact) appears on your stream. IRL backpacks make use of variable bitrate encoding which prioritizes the quality of complex parts of the scene, often resulting in blocking artifacts appearing in less detailed areas such as the sky. The adaptive deblocker is optimized for this case, where it attempts to lessen blocking artifacts while keeping all the detail in complex areas.

Demonstration Video: https://youtu.be/LCrVeJ-H1IY

Specifications

  • Filter Type: Video or effects filter
  • HDR Support: No
  • Performance Impact: Medium (>1ms)
  • Special Requirements:
    • Must have an OpenCL capable GPU.
    • Should be the first filter applied.
    • Must be applied before HUD elements (donations, subscriptions etc.)

Settings

  • Strength: The strength determines the threshold between low and high detailed parts of the scene. A higher strength allows the filter to handle more significant blocking artifacts but may lead to unintentional blurring of less prominent scenery.

  • Test Mode: Aids in testing by drawing the following information to the screen:

    • Frame Time: The amount of time required to filter a single frame on your computer. To avoid stream lag, this number should be well below 16ms for a 60FPS stream and 33ms for a 30FPS stream.

    • Deblocking Selection: The low detail regions selected for deblocking. This lets you observe how the choice of strength affects your stream.

Best Practices

The strength setting should be set to the highest value that doesn't cause undesired blurring or detail loss across the expected scenery of the stream. The default setting of 3 makes a good compromise between de-blocking performance and minimization of blurring. Some people may find that their skin is classified as low detail by the filter. This can lead to a small skin beautification effect that may not be desirable.