New test sequence and checkpoints + learnt US/DS blocks in CollapseNet (!9) · Merge requests · Kleidi / kleidi-examples / Ai Camera Pipelines · GitLab

You need to sign in or sign up before continuing.

Tarush Bansal requested to merge dev into main Oct 08, 2025

Added new test sequence with more motion to serve as better reference point for spatio-temporal vs spatial denoiser modes.
Modified both CollapseNet python and c++ implementations to use learnt US/DS blocks using conv2d/conv2d transpose operations (better IQ).
Added best checkpoints based on july training runs for both spatial-temporal and spatial denoiser modes (sigma=0.0609).