From 9dd123e1d25325f09167134e98c04fd6d7e91970 Mon Sep 17 00:00:00 2001 From: skal Date: Thu, 12 Feb 2026 12:23:17 +0100 Subject: Update docs and help messages for CNN v2 completion Updated: - HOWTO.md: Complete pipeline, storage buffer, --validate mode - TODO.md: Mark CNN v2 complete, add QAT TODO - PROJECT_CONTEXT.md: Update Effects status - CNN_V2.md: Mark complete, add storage buffer notes - train_cnn_v2_full.sh: Add --help message All documentation now reflects: - Storage buffer architecture - Binary weight format - Live training progress - Validation-only mode - 8-bit quantization TODO --- TODO.md | 42 +++++++++++++++++++++--------------------- 1 file changed, 21 insertions(+), 21 deletions(-) (limited to 'TODO.md') diff --git a/TODO.md b/TODO.md index 39b6857..a1ee9a2 100644 --- a/TODO.md +++ b/TODO.md @@ -24,33 +24,33 @@ Self-contained workspaces for parallel demo development. --- -## Priority 2: CNN v2 - Parametric Static Features (Task #85) [READY FOR TRAINING] +## Priority 2: CNN v2 - Parametric Static Features (Task #85) [COMPLETE] Enhanced CNN post-processing with multi-dimensional feature inputs. **Design:** `doc/CNN_V2.md` **Status:** -- ✅ Phase 1: Static features shader (RGBD + UV + sin encoding + bias → 8×f16, 3 mip levels) -- ✅ Phase 2: C++ effect class (CNNv2Effect skeleton, multi-pass architecture) -- ✅ Phase 3: Training pipeline (`train_cnn_v2.py`, `export_cnn_v2_shader.py`) -- ✅ Phase 4: Validation tooling (`scripts/validate_cnn_v2.sh`) -- ✅ Phase 5: Render pipeline (compute passes, bind groups, texture management) - -**Implementation complete:** -- Static features compute pass functional -- Multi-pass architecture ready -- Layer shader integration structure in place -- All tests passing (36/36) - -**Next:** Train model, generate layer shaders, integrate into demo - -**Key improvements over v1:** -- 7D static feature input (vs 4D RGB) -- Per-layer configurable kernels (1×1, 3×3, 5×5) -- Float16 weight storage (~6.4 KB) - -**Target:** <10 KB for 64k demo constraint +- ✅ Phase 1-5: All implementation phases complete +- ✅ Storage buffer architecture (dynamic layer count support) +- ✅ Binary weight format (header + layer info + f16 weights) +- ✅ Training pipeline with live progress display +- ✅ Complete validation tooling (`train_cnn_v2_full.sh --validate`) +- ✅ All tests passing (36/36) + +**Features:** +- 7D static feature input (RGBD + UV + sin encoding + bias) +- Storage buffer weights (~3.2 KB for 3-layer, 8→4→4 config) +- Dynamic layer count (not hardcoded) +- Single compute shader with per-layer params +- Patch-based training (harris detector, 32×32 patches) +- Fast training config: 100 epochs, 3×3 kernels + +**Performance:** +- Storage buffer overhead: ~10-20% vs constants (negligible @ 60fps) +- Target achieved: <10 KB for 64k demo constraint + +**TODO:** 8-bit quantization for 2× size reduction (~1.6 KB). Requires QAT. --- -- cgit v1.2.3