demo.git - Vide-coded 64k demo system

Age	Commit message (Expand)	Author
22 hours	format .wgsl layer code (cosmetics)	skal
31 hours	fix: Use patch-based inference to match CNN training distribution	skal
31 hours	opt: Move invariant in1 calculation outside CNN convolution loops	skal
32 hours	opt: Vec4-optimize CNN convolution shaders for SIMD	skal
33 hours	chore: Update CNN architecture to 3×3×3 with new trained weights	skal
33 hours	docs: Update CNN training documentation with patch extraction	skal
33 hours	feat: Add salient-point patch extraction for CNN training	skal
34 hours	fix: Correct UV coordinate computation to match PyTorch linspace	skal
34 hours	fix: Add clamp to CNN final layer to match PyTorch training	skal
35 hours	refactor: Optimize CNN grayscale computation	skal
35 hours	update train_cnn.py and shader	skal
36 hours	feat: Add inference mode to train_cnn.py for ground truth generation	skal
36 hours	fix: CNN training normalization pipeline consistency	skal
37 hours	refactor: Optimize CNN normalization to eliminate redundant conversions	skal
38 hours	fix: Support variable kernel sizes in CNN layer generation	skal
39 hours	feat: CNN RGBD→grayscale with 7-channel augmented input	skal
40 hours	udpate	skal
43 hours	feat: Add multi-layer CNN support with framebuffer capture and blend control	skal
45 hours	docs: Update and streamline CNN training documentation	skal
45 hours	feat: Add checkpointing support to CNN training script	skal
45 hours	fix: Auto-expand single kernel size to all layers in training script	skal
45 hours	update target images	skal
45 hours	feat: Add coordinate-aware CNN layer 0 for position-dependent stylization	skal