b471b84fd4
* craft-raw: the file-format layer of camera raw files, with DNG and CRW A new crate for #587 and #778: bytes in, the sensor data out as a neutral RawImage (samples, CFA, levels, active area, crop, orientation, colour tags, opcode lists, profile tables), and back for DNG. One Cargo feature per format: dng (read and write), jxl (JPEG XL DNG), crw (through craft-crw). The base is LightCraft's crates/raw at 5ff567a4: dng, dngwrite, tiffraw, unpack, ljpeg, jxl, semantic, preview, the parsing half of opcodes, profile and gaintable, the model and probe. Demosaicing and everything after it stay with the develop pipeline. Ported from PhotoCraft's and FilmCraft's DNG readers, each with a test: the default crop's per-tag defaults, an unusable crop falling back to the active area, non-finite white levels, the white default of a linearized image, the AsShotWhiteXY range, 8-bit-only lossy JPEG, lossy tiles that must cover their part of the image, and the CinemaDNG frame tags. * craft-raw: Hasselblad 3FR and FFF (feature hasselblad) RawFormat::CfaTiff decodes: the DNG-layout reader plus what these files add. Uncompressed data or Hasselblad's lossless JPEG (selection value 8), RGGB when no CFAPattern is stated, an FFF's colour matrix from the 3FR header it embeds, and the 8-bit RGB preview handed on as a baseline TIFF. Moved from LightCraft's vendor/hasselblad.rs at 5ff567a4 with its tests. Ported from PhotoCraft's reader: four BlackLevel values without a BlackLevelRepeatDim are one per CFA site in a file that is not a DNG. * craft-raw: Minolta MRW, Epson ERF, Mamiya MEF and Kodak DCS520C Four formats only LightCraft decodes, one feature each (mrw, erf, mef, kodak), moved from its vendor modules at 5ff567a4 with their tests; the MRW block container comes from lightcraft-meta. src/shared.rs holds the level helpers the maker decoders share. ERF and MEF state no white balance: the estimate LightCraft made from the preview is develop-side work, so wb_multipliers is None for them here. * craft-raw: shared scan budget for tagged previews, tile-sized lossy JPEG budget, bound on Deflate tile padding Ported from storytold/lightcraft#3695. Candidate preview ranges are deduplicated and charged to one byte budget and one header-walk budget; a lossy JPEG decoded for a tile may hold no more samples than the tile rounded up to whole MCUs (and never more than Limits::max_lossy_tile_samples); a Deflate chunk whose nominal size is mostly padding outside the image is refused before it is inflated. * craft-raw: lossless JPEG sample bound per referenced byte; previews larger than the file are passed over Ported from storytold/lightcraft#3910 (the parts whose code is in craft-raw today). check_image refuses a lossless JPEG image that declares more than 32 samples per referenced byte, counted once per chunk that points at the bytes so identical tiles stored once still decode. The Hasselblad RGB preview search passes over an image whose pixel bytes exceed the file and still takes a smaller complete one. * craft-raw: test for the Deflate padding bound * craft-raw: Pentax PEF and Samsung SRW Moved from LightCraft crates/raw (vendor/pef.rs, vendor/srw.rs) at 5ff567a4. The Huffman difference decoder and bit reader that PEF shares with the Nikon, Panasonic and Sony codings become the huffman module. * craft-raw: Canon CR2, with sRAW and mRAW Moved from LightCraft crates/raw (vendor/cr2.rs, vendor/cr2/sraw.rs) at 5ff567a4. The white-balance presets a CR2 lists come back as data in ColorData::kelvin_presets (the ordering from LightCraft's kelvin_presets.rs; the temperature and tint arithmetic stays with the developer). * craft-raw: an sRAW preview larger than the file is refused before its buffer is reserved Ported from storytold/lightcraft#3910 (cr2::sraw::read_preview). * WIP craft-raw: Nikon NEF/NRW (does not compile yet: limits threading, develop() calls in tests) * craft-raw: Nikon NEF/NRW, with PhotoCraft's maker-note compression routing and 12-bit D4 white level * craft-raw: Panasonic RW2 / RWL / RAW, every raw format * craft-raw: RW2 format 8 refuses strips stating fewer bits than samples Ported from storytold/lightcraft#3910. * craft-raw: RW2 format 8 refuses samples above the declared bit depth Ported from storytold/photocraft#11838. * craft-raw: the RW2 format 8 robustness seed carries a valid stream Its random payload decoded only while out-of-depth samples were clamped; mutations now start from a file that decodes. * craft-raw: Fujifilm RAF, uncompressed and compressed, with lens corrections * craft-raw: compressed RAF stripes cap their capacity hint at the stripe's bits Ported from storytold/lightcraft#3910. * craft-raw: test RAF orientation and camera from the embedded JPEG's Exif * craft-raw: RAF embedded preview, and its range from the header alone (bounded_preview) --------- Co-authored-by: radialmonster <radial@gmail.com> Co-authored-by: wnanbei <wnanbei@gmail.com>
6.4 KiB
Fujifilm compressed RAF
The derivation record of src/rafc.rs, moved from LightCraft (docs/raf-compression.md) with the decoder. "Supplied" files were supplied to LightCraft and are not public; the CC0 ones are pinned in raf-corpus.sha256 (paths as in LightCraft's corpus).
Format and implementation
src/raf.rs reads the RAF header, raw TIFF IFD, crop, CFA, black level and white balance. The big-endian compression field at byte 108 distinguishes uncompressed (0), lossless compressed (2) and lossy compressed (3); testing only the strip's first two bytes would misclassify an uncompressed 16-bit sample of 0x5349.
src/rafc.rs handles the compression stream:
- A 16-byte
ISheader carries compression mode, sensor type (Bayer or X-Trans), bit depth, dimensions, padded width, 768-pixel stripe width, stripe count and six-row group count. It must agree with the TIFF dimensions and CFA. - A big-endian size table, padded to 16 bytes, locates independent entropy streams. Lossy files add a table of one quantizer byte per six-row group for each stripe, each table padded to 16 bytes.
- Each six-row group is collated into three red, six green and three blue vectors. Bayer vectors have 384 samples; X-Trans vectors have 512 slots, including interpolation slots that consume no bits. The last stripe's padding is decoded for predictor state but excluded from the output.
- Colour-vector pairs alternate R/G and G/B. Even columns lead odd columns by seven positions. Two preceding lines of each colour and the preceding line's edge values provide prediction context.
- Signed prediction errors use adaptive Rice coding: a unary prefix followed by a context-dependent number of bits, with a bounded escape for large errors. Three pair phases have separate even/odd statistics. The 81 signed gradient contexts fold into 41 states.
- Lossy quantizers change the error range and reconstruction step. Low-gradient sites retain finer precision (0, 1 or 2), with five folded contexts per precision and statistics that survive main-quantizer changes. The main context statistics reset when its quantizer changes. Their initial sum is
max(2, (range + 32) / 64); using 16 as the minimum breaks more heavily quantized files.
Stripes decode in parallel using the existing Rayon dependency. The entropy reader keeps up to 64 bits buffered and counts a unary prefix's zeros in one step instead of bit by bit; a property test checks it against reading bit by bit, errors included. Offsets, sizes, dimensions and entropy reads are checked; truncated data, impossible codes and inconsistent headers return errors. Unknown container compression modes, disagreements between container and stream modes, and compressed X-Trans layouts other than the supported CFA return errors. Header-only probing validates the tables and stripe ranges without decompressing the image.
Verification
Complete sensor arrays were compared sample by sample with an external decoder used only as a black-box oracle (rawpy 0.27.1). No decoder source, Adobe data, camera matrices or profiles were used. This tool is not a product or test dependency.
| Coverage | Files |
|---|---|
| Supplied X-H2S | DSCF4192 uncompressed; DSCF4193 lossless; DSCF4194 lossy |
| Supplied X-T4 | _DSF2355 / _DSF2356 uncompressed; _DSF2357 lossless; _DSF2358 lossy |
| CC0 X-Trans | X-T2, X-T20, X-T4, X-H2, X-T5, X-M5, X-T50, X-E5 |
| CC0 Bayer | GFX 50S (14-bit), GFX 100 (14/16-bit), GFX100S (16-bit lossless/lossy), GFX100RF (16-bit lossy) |
All 24 files matched the reference sensor arrays exactly. This covers 13 camera bodies and both older and current compression variants. Synthetic tests also cover 12-bit streams; a real compressed 12-bit camera file has not been verified. Support is selected from the file's layout, without a camera-name allowlist. Other bodies using this layout should decode, but this is not a claim that every Fujifilm model has been tested. Older FinePix layouts without the supported raw TIFF IFD remain unsupported.
The 17 CC0 files live in the gitignored corpus. Their published SHA-256 identities are pinned in raf-corpus.sha256; their source licence records were checked as CC0 1.0. cargo xtask corpus --download fetches them. cargo test -p lightcraft-raw corpus_fujifilm_compressed_samples -- --nocapture verifies each available file's checksum before checking its reference sensor sum and position-weighted sum. The test skips absent corpus files; the full-array comparisons were performed locally. No supplied media or large binary fixtures are committed.
Unconditional procedural tests cover Bayer/X-Trans, 12/14/16 bits, partial final stripes, quantizer changes, residual signs and escapes, malformed headers and truncated/random entropy streams. Regressions also check overflowing uncompressed row strides, unsupported modes and CFA layouts, and container/stream mode mismatches.
Local release decode measurements: about 0.16–0.24 seconds for 24–40 MP X-Trans files, 0.29 seconds for 50 MP Bayer, and 0.5–0.6 seconds for 100 MP Bayer. These measure sensor decoding, not the full develop/export pipeline. The buffered prefix count took 13–17 % off the decode time on an Apple M2 (8 cores, minimum of 5 runs, old and new build alternated): X-T2 178 → 152 ms, X-S10 lossless 265 → 226 ms, X-E5 442 → 385 ms, X-H2 448 → 378 ms, GFX 50S 522 → 435 ms, GFX 100 762 → 644 ms, with sensor arrays unchanged bit for bit.
Sources
- Fabian's original prose description of lossless RAF compression: independent stripes, colour vectors, iteration order, predictors' neighbourhoods, Rice coding and gradient adaptation. Its linked implementations were not consulted.
- ExifTool Fujifilm tag-name documentation: RAF header and raw-IFD metadata fields.
- Fujifilm's compression-mode explanation.
- raw.pixls.us: CC0 camera samples, published checksums and licence metadata.
- rawpy's documented
raw_imageAPI: the complete sensor array includes margins, allowing comparisons before demosaic or colour processing.
Exact byte layouts, colour-vector masks, prediction formulas, adaptation parameters and lossy precision selection were established through sample analysis and black-box comparisons.