demo.git/workspaces/main/shaders/cnn, branch main

archive(cnn): move CNN v1 to cnn_v1/ subdirectory

2026-02-15T17:52:48Z

Consolidate CNN v1 (CNNEffect) into dedicated directory: - C++ effect: src/effects → cnn_v1/src/ - Shaders: workspaces/main/shaders/cnn → cnn_v1/shaders/ - Training: training/train_cnn.py → cnn_v1/training/ - Docs: doc/CNN*.md → cnn_v1/docs/ Updated all references: - CMake source list - C++ includes (relative paths: ../../cnn_v1/src/) - Asset paths (../../cnn_v1/shaders/) - Documentation cross-references CNN v1 remains active in timeline. For new work, use CNN v2 with enhanced features (7D static, storage buffer, sigmoid activation). Tests: 34/34 passing (100%)

add trained layers

2026-02-11T23:10:36Z

+misc

docs: Update CNN comments and add bias fix summary

2026-02-11T22:25:09Z

- Fix stale comments: RGBD→RGB (not grayscale) - Clarify shape transformations in inference - Add CNN_BIAS_FIX_2026-02.md consolidating recent fixes - Include regenerated weights with 5x5 kernel for layer 0 Co-Authored-By: Claude Sonnet 4.5

fix: CNN bias accumulation and output format improvements

2026-02-11T22:13:43Z

- Fix bias division bug: divide by num_positions to compensate for shader loop accumulation (affects all layers) - train_cnn.py: Save RGBA output preserving alpha channel from input - Add --debug-hex flag to both tools for pixel-level debugging - Remove sRGB/linear_png debug code from cnn_test - Regenerate weights with corrected bias export Co-Authored-By: Claude Sonnet 4.5

update cnn code

2026-02-11T16:23:07Z

fix: Register cnn_conv1x1 snippet and add verification

2026-02-11T15:06:16Z

- Add cnn_conv1x1 to shader composer registration - Add VerifyIncludes() to detect missing snippet registrations - STRIP_ALL-protected verification warns about unregistered includes - Fixes cnn_test runtime failure loading cnn_layer.wgsl Co-Authored-By: Claude Sonnet 4.5

fix: Move sigmoid activation to call site in CNN layer shader

2026-02-11T09:10:11Z

Conv functions now return raw sum, sigmoid applied at call site. Matches tanh pattern used for inner layers. Co-Authored-By: Claude Sonnet 4.5

fix: Replace clamp with sigmoid in CNN final layer

2026-02-11T09:07:58Z

Final layer used hard clamp causing saturation to white when output > 1.0. Replaced with sigmoid activation for smooth [0,1] mapping with gradients. Changes: - train_cnn.py: torch.sigmoid() in forward pass and WGSL codegen - WGSL shaders: 1.0/(1.0+exp(-sum)) in cnn_conv3x3/5x5 _7to1 functions Co-Authored-By: Claude Sonnet 4.5

format .wgsl layer code (cosmetics)

2026-02-11T08:18:18Z

opt: Move invariant in1 calculation outside CNN convolution loops

2026-02-10T23:26:25Z

The in1 vector (uv_norm, gray, 1.0) is loop-invariant and doesn't depend on dx/dy offset. Moving it outside the convolution loop eliminates redundant computation and enables better SIMD optimization. Updated both shader files and train.py code generation. Co-Authored-By: Claude Sonnet 4.5