demo.git - Vide-coded 64k demo system

Age	Commit message (Expand)	Author
11 hours	Refine training script output and validation	skal
11 hours	TODO: 8-bit weight quantization for 2× size reduction	skal
11 hours	CNN v2: Storage buffer complete - real weights exported	skal
11 hours	CNN v2: storage buffer architecture foundation	skal
12 hours	TODO: Add random sampling to patch-based training	skal
12 hours	CNN v2: Patch-based training as default (like CNN v1)	skal
12 hours	Fix: CNN v2 training - handle variable image sizes	skal
12 hours	CNN v2: parametric static features - Phases 1-4	skal
14 hours	remove more stale files	skal
23 hours	feat: implement beat-based timing system	skal
23 hours	add trained layers	skal
24 hours	docs: Update CNN comments and add bias fix summary	skal
24 hours	fix: CNN bias accumulation and output format improvements	skal
30 hours	update cnn code	skal
31 hours	refactor: Use linspace(-1,1) directly for coords	skal
31 hours	fix: Compute gray from [0,1] RGB in CNN shader generator	skal
35 hours	fix: Complete auxiliary texture initialization fix	skal
37 hours	add --save-intermediates to train.py and cnn_test	skal
37 hours	fix: Move sigmoid activation to call site in CNN layer shader	skal
37 hours	fix: Replace clamp with sigmoid in CNN final layer	skal
38 hours	feat: Add early stopping to CNN training	skal
38 hours	fix: CNN training/inference to match WGSL sliding window	skal
38 hours	format .wgsl layer code (cosmetics)	skal
47 hours	fix: Use patch-based inference to match CNN training distribution	skal
47 hours	opt: Move invariant in1 calculation outside CNN convolution loops	skal
2 days	opt: Vec4-optimize CNN convolution shaders for SIMD	skal
2 days	chore: Update CNN architecture to 3×3×3 with new trained weights	skal
2 days	docs: Update CNN training documentation with patch extraction	skal
2 days	feat: Add salient-point patch extraction for CNN training	skal
2 days	fix: Correct UV coordinate computation to match PyTorch linspace	skal
2 days	fix: Add clamp to CNN final layer to match PyTorch training	skal
2 days	refactor: Optimize CNN grayscale computation	skal
2 days	update train_cnn.py and shader	skal
2 days	feat: Add inference mode to train_cnn.py for ground truth generation	skal
2 days	fix: CNN training normalization pipeline consistency	skal
2 days	refactor: Optimize CNN normalization to eliminate redundant conversions	skal
2 days	fix: Support variable kernel sizes in CNN layer generation	skal
2 days	feat: CNN RGBD→grayscale with 7-channel augmented input	skal
2 days	udpate	skal
2 days	feat: Add multi-layer CNN support with framebuffer capture and blend control	skal
3 days	docs: Update and streamline CNN training documentation	skal
3 days	feat: Add checkpointing support to CNN training script	skal
3 days	fix: Auto-expand single kernel size to all layers in training script	skal
3 days	update target images	skal
3 days	feat: Add coordinate-aware CNN layer 0 for position-dependent stylization	skal