Expose crop dimensions

[jtomori]

Hello, this is my attempt at exposing cropped dimensions, aiming to solve #139.

ImageSizes gets new fields: crop_left_margin, crop_top_margin, crop_width, crop_height matching LibRaw's raw_inset_crops.

In Canon CR2s this corresponds to the intended image resolution. E.g. 5616x3744 in RAW_CANON_5DMARK2_PREPROD.CR2, vs 5634x3752 which is in the width, height fields:

ImageSizes(
    raw_height=3804, raw_width=5792, 
    height=3752, width=5634, 
    top_margin=52, left_margin=158, 
    iheight=3752, iwidth=5634, 
    pixel_aspect=1.0, flip=0, 
    crop_left_margin=168, crop_top_margin=56, 
    crop_width=5616, crop_height=3744)

Exiftool and exiv2 report 5616x3744 too.

In case of the test RAW files from Nikon, Sigma, Kodak all new fields are set to 0.

[jtomori]

However when I've checked the new crop fields for a0001-jmac_DSC1459.dng (from #139), I see the following results, which are off by 1 pixel:

ImageSizes(
    raw_height=2014, raw_width=3040, 
    height=2014, width=3040, 
    top_margin=0, left_margin=0, 
    iheight=2014, iwidth=3040, 
    pixel_aspect=1.0, flip=0, 
    crop_left_margin=16, crop_top_margin=8, 
    crop_width=3007, crop_height=1999)

Exiftool:

[TIFF-IFD0-SubIFD] DefaultCropOrigin            : 15 7
[TIFF-IFD0-SubIFD] DefaultCropSize              : 3008 2000

I am surprised that I've got the expected values in the case of the CR2 above, but get off-by-one values in case of this DNG.

In the added code I'm just exposing LibRaw's values. Could LibRaw return incorrect values in case of DNG? 🤔

[jtomori]

I've just tried converting RAW_CANON_5DMARK2_PREPROD.CR2 to DNG with Adobe DNG Converter (17.1.0) and got this ImageSizes from the DNG:

ImageSizes(
    raw_height=3804, raw_width=5792, 
    height=3752, width=5634, 
    top_margin=52, left_margin=158, 
    iheight=3752, iwidth=5634, 
    pixel_aspect=1.0, flip=0, 
    crop_left_margin=336, crop_top_margin=112, # This differs from what I've got from CR2, for some reason it's *2 here
    crop_width=5616, crop_height=3744)

Exiftool of the DNG:

[TIFF-IFD0-SubIFD] DefaultCropOrigin            : 10 5
[TIFF-IFD0-SubIFD] DefaultCropSize              : 5616 3744

Hmm, not sure what's going on here.

[letmaik]

Do you want to experiment a bit more?

[jtomori]

I think I am out of ideas and it's working for my use-case: CR2s.

As I am just exposing what LibRaw is providing I fear that there's not much I can do.

At first I was optimistic about seeing nice results on CR2 and test data from other cameras, but that DNG with off by one error reduced my confidence in this patch.

[letmaik]

It might be worth posting on the libraw forum, maybe it's an actual bug?

[jtomori]

Good idea, will try accessing those crop values from LibRaw's CLI programs to compare with the ones I get in rawpy. Depending on findings it might be worth flagging on LibRaw forum.

[jtomori]

I've checked cropped dimensions as reported by LibRaw against those exposed by this patch and they all match.

I've used LibRaw's (0.21.3) raw-identify -v and looked for line starting with Raw inset, width x height:

• RAW_CANON_5DMARK2_PREPROD.CR2
  • 5616 x 3744 left: 168 top: 56
• a0001-jmac_DSC1459.dng
  • 3007 x 1999 left: 16 top: 8
• RAW_CANON_5DMARK2_PREPROD.dng
  • 5616 x 3744 left: 336 top: 112

I'd say that this patch is working as intended - it's exposing values provided by LibRaw. Do you think this is good to be merged?

I will ask on LibRaw's forum about the off by one error and the DNG as it looks to be a bug on their side.

EDIT: Corresponding forum post for reference https://www.libraw.org/node/2838

[jtomori]

Summary of replies from LibRaw forum: the off-by-one difference in the DNG is expected behavior caused by alignment to bayer pattern.

The DNG left, top padding difference was a bug and has been fixed with LibRaw/LibRaw@29d9785, and will be probably included in the next LibRaw release.

I've increased vcpkg's version in the Windows build script which should fix the previously failing Windows CI builds.

[letmaik]

Looks good, great this even surfaced a bug in libraw! I'll merge it once CI ran through.