demo.git - Vide-coded 64k demo system

Age	Commit message (Expand)	Author
23 hours	Replace hard clamp with sigmoid activation in CNN v2	skal
24 hours	Fix CNN v2 training: always save final checkpoint, derive num_layers	skal
26 hours	CNN v2 training: Fix float64/float32 dtype mismatch in depth feature	skal
27 hours	CNN v2: Use alpha channel for p3 depth feature + layer visualization	skal
32 hours	CNN v2 training: Add --grayscale-loss option for luminance-based loss computa...	skal
32 hours	CNN v2: Change feature #6 from sin(10x) to sin(20y)	skal
32 hours	CNN v2: Add TODO for flexible feature layout in binary format v3	skal
33 hours	CNN v2: Add --mip-level option for parametric features	skal
37 hours	CNN v2 training: Use target image alpha channel	skal
37 hours	CNN v2: Restore per-layer kernel sizes support	skal
37 hours	CNN v2: Refactor to uniform 12D→4D architecture	skal
3 days	Refine training script output and validation	skal
3 days	TODO: 8-bit weight quantization for 2× size reduction	skal
3 days	TODO: Add random sampling to patch-based training	skal
3 days	CNN v2: Patch-based training as default (like CNN v1)	skal
3 days	Fix: CNN v2 training - handle variable image sizes	skal
3 days	CNN v2: parametric static features - Phases 1-4	skal