Skip to content

v2.1
Compare
Choose a tag to compare
@lnstadrum lnstadrum released this 19 Apr 19:34
· 41 commits to master since this release
v2.1
845bb62
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Features:

  • Introducing Raspberry Pi 4 support
    • USE_EGL_DRM=ON needs to be supplied to cmake when compiling for Raspberry Pi 4
    • For Raspberry Pi prior to 4 series, the cmake options become USE_EGL=ON and USE_BRCM_LIBS=ON
  • Updating X2 neural net and implementation
    • 32.64 dB PSNR on DIV2K validation set, previously 32.57 dB
    • Reducing register pressure in shaders
  • Introducing dog classification neural net featuring inference with NNets
    • 72% top-1 validation accuracy on 120-class subset of ILSVRC 2012 containing images of dogs with a model of 225k parameters
    • Introducing a piecewise sigmoid-like activation suitable for the shader-based inference implementation
  • Supporting clipped ReLU activation
  • Introducing GLSL dialect for better compatibility with different GPUs
    • Automatically transforms the shader code into GLSL ES 2.0, GLSL ES 3.1 or GLSL 1.3 depending on the OpenGL version supported by the GPU
  • Fixing bugs and introducing new ones

Prebuilt Python packages

Prebuilt Python packages are available for download below:

  • for Python 3.5, 3.6, 3.7, 3.8, 3.9 for Ubuntu 20.04 (x64),
  • for Python 3.6 for Nvidia® Jetson™ / JetPack 4.3 (aarch64) .

Prebuild Python packages for Windows can be installed with pip install beatmup.

Assets 8
Loading