aleleba/lightcraft
1
0
Fork 0
mirror of https://github.com/storytold/lightcraft.git synced 2026-10-24 10:09:01 -06:00
Code Issues Packages Projects Releases Wiki Activity

LR-EDIT-DETAIL-DENOISE: address installation and model-policy review

Browse Source 
Generated with Codex.
This commit is contained in:
Daniel Voshart
2026-10-08 22:07:23 -04:00
parent 7935999d94
commit fd4019300a
59 changed files with 1437 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/denoise.md
+22 -9
View File
@@ -15,22 +15,29 @@ An active slider gesture finishes before the pending action resumes.
## Models and terms
**No model weights are bundled.** Models live in the per-user `denoise-models` folder
(override with `LIGHTCRAFT_DENOISE_MODELS`). Settings offers a pinned, verified download
of RawNIND UtNet2 for Bayer sensors, or installation of a user-supplied ONNX file with
a `denoise-model.json` beside it. Terms must be acknowledged before installation or
(override with `LIGHTCRAFT_DENOISE_MODELS`). Settings provides **Install from file…**
for a user-supplied ONNX file with a `denoise-model.json` beside it. The default build
offers no model downloads. The separate `rawnind-model` Cargo feature enables the
pinned RawNIND offer only when a release's model policy permits it.
Terms must be acknowledged before installation or
download. Downloads use the shared pure-Rust `lightcraft-fetch` client, with resume,
progress, cancellation, size limits and SHA-256 verification.
RawNIND's offered ONNX weights are **GPL-3.0**, as recorded in the pinned model catalog.
RawNIND's optional ONNX weights are **GPL-3.0**, as recorded in the pinned model catalog.
They are downloaded separately after consent and read by our own Rust executor.
Whether releases should offer this download remains a maintainer policy decision;
permissively licensed or independently trained weights remain a gap. This is not a
claim that optional downloading settles licensing questions. No GPL decoder or neural
runtime code is included. Model metadata is in `crates/denoise/src/known.rs`.
runtime code is included. Model metadata is in `crates/denoise-core/src/known.rs`.
Installation checks the manifest, model hash and size, then runs a synthetic self-test:
finite output of the promised shape, repeatability, the declared scale and reduced
noise. Unsupported operators, malformed files and a failed self-test return errors.
The UI performs copying, extraction, hashing, parsing and CPU/GPU self-tests on one
background installation worker. `denoise.models.downloads` and `denoise.pump` report
`installing`, `installed` or `failed`; an existing model is preserved on a failed replacement.
Command clients can request the same asynchronous path with `background: true`.
The synchronous command remains available for scripts that require the result immediately.
The current contract is `bayerToRgb`: normalized, black-subtracted, non-white-balanced
RGGB planes `[1, 4, tile, tile]` produce camera RGB `[1, 3, 2*tile, 2*tile]`.
@@ -63,10 +70,16 @@ reduction and report their limitation.
## Pure-Rust CPU and GPU execution
The CPU executor reads a bounded subset of ONNX into checked network data. Convolutions
use bounded im2col blocks and `faer-core 0.17.1` with its Rust-intrinsic GEMM kernels,
without BLAS, generated assembly or an inner thread pool. Tiles share the outer worker
use bounded im2col blocks and `faer-core 0.17.1` with its pure-Rust GEMM backend,
without BLAS, C dependencies or an inner thread pool. That backend includes
architecture-specific assembly in its dependencies. Tiles share the outer worker
pool. A scalar implementation checks the optimized operators on synthetic networks.
Tract is not a workspace dependency or a product feature.
CPU inference and the GPU model module are feature-gated; ordinary engine/web builds
do not compile the faer/GEMM backend. The engine depends on the inference crate only through `denoise = ["dep:lightcraft-denoise", …]`;
lightweight `lightcraft-denoise-core` metadata and cache formats remain available without inference. Pictures are capped at 100 megapixels; at most
four tile calls run together, chosen within the process-wide working-memory budget.
One idle CPU workspace is retained, and cache costs reserve room for the blended picture.
The GPU executor uses our WGSL kernels through wgpu. It checks a real model's output
against the CPU and falls back on errors, unsupported networks or a slower device.
@@ -97,8 +110,8 @@ Fresh checks can be run using an already installed model and CC0 raw samples:
```text
LC_DENOISE_MODEL=<model.onnx> cargo test --release -p lightcraft-gpu --lib real_model_on_the_gpu_matches_cpu -- --ignored --nocapture
LC_DENOISE_MODEL=<model.onnx> LC_DENOISE_RAW=<corpus/raw> cargo test --release -p lightcraft-engine --features denoise --test denoise_cpu -- --ignored --nocapture
LC_DENOISE_MODEL=<model.onnx> LC_DENOISE_RAW=<corpus/raw> cargo test --release -p lightcraft-engine --features denoise --test denoise_real -- --ignored --nocapture
LC_DENOISE_MODEL=<model.onnx> LC_DENOISE_RAW=<corpus/raw> cargo test --release -p lightcraft-engine --features rawnind-model --test denoise_cpu -- --ignored --nocapture
LC_DENOISE_MODEL=<model.onnx> LC_DENOISE_RAW=<corpus/raw> cargo test --release -p lightcraft-engine --features rawnind-model --test denoise_real -- --ignored --nocapture
```
## Commands and limits
docs/parity.md
+2 -2
View File
@@ -26,7 +26,7 @@ and prints the summary; `cargo xtask parity --write` refreshes the summary table
| C. Views & navigation (VIEW) | 15 | 1 | 1 | 0 | 8/9 (89%) | 4/4 (100%) |
| D. Search & filter (FILT) | 11 | 1 | 1 | 0 | 4/4 (100%) | 4/4 (100%) |
| E. Metadata (META) | 5 | 1 | 0 | 0 | 2/2 (100%) | 2/2 (100%) |
| F. Edit panel — global adjustments (EDIT) | 42 | 1 | 5 | 1 | 28/28 (100%) | 13/14 (93%) |
| F. Edit panel — global adjustments (EDIT) | 42 | 2 | 4 | 1 | 28/28 (100%) | 13/14 (93%) |
| G. Profiles (PROF) | 7 | 1 | 3 | 0 | 3/4 (75%) | 3/3 (100%) |
| H. Crop & rotate (CROP) | 9 | 0 | 0 | 1 | 6/6 (100%) | 3/3 (100%) |
| I. Remove / healing (REM) | 7 | 1 | 2 | 2 | 4/4 (100%) | 2/3 (67%) |
@@ -79,7 +79,7 @@ Take the first one nobody is working on.
Object and Describe masks run SAM 3 in pure Rust (`crates/segment`); the weights (SAM License) are never bundled:
the app offers a consented, verified download, but **LightCraft's CDN mirrors are not configured yet** (the
built-in list is empty, so today users need their own mirror or a manual install). Subject / Sky / People could
use the same model with fixed prompts. Bayer RAW denoise now runs on pure-Rust CPU/GPU with separately installed weights; its opt-in GPL model policy, linear RGB and quality comparison remain open. Super-resolution models remain a maintainer decision.
use the same model with fixed prompts; denoise / super-resolution models remain a maintainer decision.
7. **HDR** (Q. HDR, LR-EXP-HDR), **video** (R. Video), **Classic output modules** (Map view, Book, Slideshow module,
Print, publish): large, well understood, lower priority than 1–5.
8. **LR-VIEW-ZOOM** (P0): pinch zoom and two-finger pan now work, with stable image cursors between gesture events.