summaryrefslogtreecommitdiff
path: root/training
AgeCommit message (Expand)Author
2026-02-27remove old files, add new training setskal
2026-02-15refactor(cnn): rename cnn_effect to cnn_v1_effect for clarityskal
2026-02-15archive(cnn): move CNN v1 to cnn_v1/ subdirectoryskal
2026-02-15refactor(cnn): isolate CNN v2 to cnn_v2/ subdirectoryskal
2026-02-14CNN v2: bilinear mip-level sampling and UI improvementsskal
2026-02-14Replace hard clamp with sigmoid activation in CNN v2skal
2026-02-14Streamline CNN v2 training pipeline outputskal
2026-02-14Fix CNN v2 training: always save final checkpoint, derive num_layersskal
2026-02-14Fix --mix option: blend prev layer with static p4-p7, not p0-p3skal
2026-02-14Fix CNN v2 static feature channel mapping (p4-p7 → channels 8-11)skal
2026-02-14gen_identity_weights: Change --mix to 50-50 blendskal
2026-02-14gen_identity_weights: Add --p47 option for static feature visualizationskal
2026-02-14gen_identity_weights: Add --mix option for static feature blendingskal
2026-02-13CNN v2: Fix Layer 0 visualization scale (was 0.5, now 1.0)skal
2026-02-13CNN v2: Add debugging tools for mismatch investigationskal
2026-02-13CNN v2 training: Fix float64/float32 dtype mismatch in depth featureskal
2026-02-13CNN v2: Alpha channel depth handling and layer visualizationskal
2026-02-13CNN v2: Use alpha channel for p3 depth feature + layer visualizationskal
2026-02-13CNN v2 training: Add --grayscale-loss option for luminance-based loss computa...skal
2026-02-13CNN v2: Change feature #6 from sin(10*x) to sin(20*y)skal
2026-02-13CNN v2: Add TODO for flexible feature layout in binary format v3skal
2026-02-13CNN v2: Add mip-level support to runtime effectskal
2026-02-13CNN v2 export: Read and display mip_level from checkpointsskal
2026-02-13CNN v2: Add --mip-level option for parametric featuresskal
2026-02-13CNN v2: Fix activation function mismatch between training and inferenceskal
2026-02-13CNN v2 training: Use target image alpha channelskal
2026-02-13CNN v2: Restore per-layer kernel sizes supportskal
2026-02-13CNN v2: Refactor to uniform 12D→4D architectureskal
2026-02-13Add weights/ subdirectory to workspaces for CNN training outputsskal
2026-02-12test_demo: Add beat-synchronized CNN post-processing with version selectionskal
2026-02-12Refine training script output and validationskal
2026-02-12TODO: 8-bit weight quantization for 2Ɨ size reductionskal
2026-02-12CNN v2: Storage buffer complete - real weights exportedskal
2026-02-12CNN v2: storage buffer architecture foundationskal
2026-02-12TODO: Add random sampling to patch-based trainingskal
2026-02-12CNN v2: Patch-based training as default (like CNN v1)skal
2026-02-12Fix: CNN v2 training - handle variable image sizesskal
2026-02-12CNN v2: parametric static features - Phases 1-4skal
2026-02-12remove more stale filesskal
2026-02-12feat: implement beat-based timing systemskal
2026-02-12add trained layersskal
2026-02-11docs: Update CNN comments and add bias fix summaryskal
2026-02-11fix: CNN bias accumulation and output format improvementsskal
2026-02-11update cnn codeskal
2026-02-11refactor: Use linspace(-1,1) directly for coordsskal
2026-02-11fix: Compute gray from [0,1] RGB in CNN shader generatorskal
2026-02-11fix: Complete auxiliary texture initialization fixskal
2026-02-11add --save-intermediates to train.py and cnn_testskal
2026-02-11fix: Move sigmoid activation to call site in CNN layer shaderskal
2026-02-11fix: Replace clamp with sigmoid in CNN final layerskal