2.1 KiB
To-Do list for Human library
Work-in-Progress
N/A
Exploring
- Optical flow for intelligent temporal interpolation
https://docs.opencv.org/3.3.1/db/d7f/tutorial_js_lucas_kanade.html - Advanced histogram equalization for optimization of badly lit scenes
Adaptive, Contrast Limited, CLAHE - TFLite models
https://js.tensorflow.org/api_tflite/0.0.1-alpha.4/ - Body segmentation with temporal analysis
https://github.com/PeterL1n/RobustVideoMatting
Known Issues
Face with Attention
FaceMesh-Landmarks
model is supported only with CPU
and WebGL
backends due to TFJS
issues
WASM
Support for WASM SIMD and WASM MultiThreading is still disabled by default in Chromium based browsers
Suggestion is to enable it manually for major performance boost
Enable via chrome://flags/#enable-experimental-webassembly-features
WebGPU
Experimental support only until support is officially added in Chromium
Enable via chrome://flags/#enable-unsafe-webgpu
Firefox
Running in web workers requires OffscreenCanvas
which is still disabled by default in Firefox
Enable via about:config
-> gfx.offscreencanvas.enabled
Face Detection
Enhanced rotation correction for face detection is not working in NodeJS due to missing kernel op in TFJS
Feature is automatically disabled in NodeJS without user impact
Pending Release Changes
- Updated FaceMesh-Landmarks models
- Added FaceMesh-with-Attention model is disabled by defauls, enable using
config.face.mesh.attention = true
- If FaceMesh-with-Attention model is enabled, Iris model gets disabled
as its functionality is superseded by attention model - Results include more detailed face mesh annotations
result.face[].annotations
- Update benchmarks:
demo/benchmark
- Slight performance improvement with canvas reads using
WASM
backend - Default to RTM version of
WASM
binaries - Enhanced
WebGPU
compatibility - Support for NodeJS v18
- Updated Known Issues
See https://github.com/vladmandic/human/blob/main/TODO.md