Skip to content

revelo v0.4.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 31 May 00:36
· 24 commits to main since this release
v0.4.0
2ebb194

Added

  • EXIF/IPTC/XMP/ICC/C2PA/MakerNotes promoted to their own stream sections
    (revelo extensions beyond MediaInfo's stream_t), matching how exiftool groups
    metadata by family-0 type rather than folding everything into General. EXIF
    streams from the container parser and the IFD walker are now merged into one.

  • IPTC IIM now anchored to the Photoshop 8BIM Image Resource Block (resource
    0x0404) instead of scanning raw bytes — fixes false-positive datasets (e.g. a
    garbage Headline) matched in compressed image data.

  • Optional exiftool-tables feature (off by default) — opt in to richer
    maker-note tag names and PrintConv value decoding sourced from ExifTool, for 14
    vendors (Apple, Canon, Casio, DJI, FLIR, FujiFilm, Minolta, Nikon, Olympus,
    Panasonic, Pentax, Samsung, Sigma, Sony). The tables live in a separate
    GPL/Artistic crate revelo-exiftool-tables (generated from ExifTool via
    codegen/extract.pl); the default revelo build stays BSD-2-Clause and uses the
    hand-written clean-room tables. Enabling the feature subjects the resulting
    binary to GPL/Artistic terms.

  • Canon maker-note decoding (under exiftool-tables) — ExifTool-derived
    ProcessBinaryData sub-tables (CameraSettings, ShotInfo) are now decoded into
    individual named, value-converted tags. Two correctness fixes make this work on
    real files: the bare-IFD offset is no longer mis-detected when the first tag id
    is 0x0000 (broke every Canon file whose maker note starts that way, e.g. the
    IXUS), and a cross-validated FixBase recovers mis-based maker-note offsets in
    edited/re-muxed files (matching exiftool's "Adjusted MakerNotes base by N").
    Records are accepted only when their self-describing byte-count header validates,
    so a wrong offset decodes nothing rather than garbage. The header-less
    FIRST_ENTRY 0 FocalLength sub-table is decoded only when FixBase confirmed the
    base. The variable-length AFInfo (0x0012) record (NumAFPoints-driven
    AFAreaXPositions[N] etc.) is hand-walked behind an exact total-size match, so a
    mis-laid record decodes nothing. The newer AFInfo2 (0x0026) / AFInfo3 (0x003C,
    SerialData) records — which lead with a self-validating AFInfoSize header and
    carry an AFAreaMode enum plus four [N] arrays — are decoded the same way. For
    Main tags revelo's older bespoke table lacks, the fuller ExifTool Main table is
    consulted under the feature. Further ProcessBinaryData sub-tables are decoded
    with per-table element size — MyColors and ContrastInfo (int16), TimeInfo and
    AspectInfo (int32), and FaceDetect3. ImageUniqueID (16-byte hex) and DateStampMode
    decode too. Validated against real camera samples: IXUS 400 31%→99%, PowerShot S40
    →99%, edited HDR files (AFInfo3, newer body) →100%, all with exact value matches.

  • EXIF / Interop / Canon Main tag names aligned to ExifTool under
    exiftool-tables for the handful of divergences (DateTime→ModifyDate,
    DateTimeDigitized→CreateDate, PhotographicSensitivity→ISO,
    PixelX/YDimension→ExifImage{Width,Height},
    Interoperability{Index,Version}→Interop{Index,Version},
    CanonOwnerName→OwnerName, CanonImageNumber→FileNumber,
    CanonThumbnailValidArea→ThumbnailImageValidArea). Lens fields are
    deliberately not remapped (the lens-formatting post-pass keys off the names).
    With all of the above, clean Canon samples reach 99% tag parity with exiftool.

Fixed

  • Maker-note parsing repaired for several vendors (BSD core, all builds). The
    Fujifilm maker note read its IFD offset from the wrong byte (12 instead of 8) and
    mis-resolved value offsets — it now parses against the full block from the correct
    offset. Nikon Type 2/3 (embedded TIFF header at offset 10) is detected so COOLPIX
    bodies parse. Header-less big-endian Konica-Minolta IFDs are now byte-order
    detected. Under exiftool-tables, Nikon names are aligned to ExifTool. Validated
    against real samples: Fujifilm 68%→98-100%, Nikon COOLPIX 63%→95%, Konica-Minolta
    76%→97%.
  • Olympus "type-2" maker notes (OLYMPUS\0II…) are now parsed: the main IFD plus
    its nested Equipment / CameraSettings / RawDevelopment / ImageProcessing / FocusInfo
    sub-IFDs (decoded via generated ExifTool tables under the feature).
  • Nikon AFInfo (0x0088) / FlashInfo (0x00A8) decoded on COOLPIX bodies.
  • UNDEFINED (type 7) values are read as a single trimmed blob instead of a string
    repeated from every byte offset (fixes e.g. Panasonic InternalSerialNumber).
  • Canon maker-note Main tag ids corrected (BSD core, all builds): 0x000C is
    the serial number (was mislabelled CanonModelID), 0x0010 is the model id
    (was CanonThumbnailValidArea), 0x0013 is the thumbnail valid area (was
    missing), 0x0015 is the serial-number format, 0x001C is DateStampMode (was
    CanonAFInfo), and 0x0028 is ImageUniqueID rendered as 16-byte hex (was
    CanonCRWParam). The model-name lookup now applies to 0x0010 instead of the
    serial number at 0x000C. Verified against exiftool's raw values.

Changed

  • Dropped the creatingLibrary header from JSON and XML output. The block
    (name/version/url) only identified the producing tool and carried nothing
    about the parsed file, so it is no longer emitted. JSON now opens directly at
    "media"; XML goes straight from the <MediaInfo> header to <media>. This is
    a deliberate divergence from mediainfo's wire format — consumers that validate
    against the published mediainfo schema or read creatingLibrary.version will see
    a missing key. revelo-diff strips the oracle's creatingLibrary line before
    comparing, so the diff harness stays focused on per-stream differences.
Assets 10
Loading