This is about READING image metadata

Writing is not supported, and never will.

I welcome PRs with fixes, but please raise an issue first if you want to add new features.

Performance

Extracting EXIF performs well, ref. the benhcmark below. Note that you can get a significant boost if you only need a subset of the fields (e.g. only the Orientation). The last line is with the library that Hugo used before it was replaced with this.

BenchmarkDecodeCompareWithGoexif/bep/imagemeta/exif/jpeg/alltags-10                62474             19054 ns/op            4218 B/op        188 allocs/op
BenchmarkDecodeCompareWithGoexif/bep/imagemeta/exif/jpeg/orientation-10           309145              3723 ns/op             352 B/op          8 allocs/op
BenchmarkDecodeCompareWithGoexif/rwcarlsen/goexif/exif/jpg/alltags-10              21987             50195 ns/op          175548 B/op        812 allocs/op

Looking at some more extensive tests, testing different image formats and tag sources, we see that the current XMP implementation leaves a lot to be desired (you can provide your own XMP handler if you want).

BenchmarkDecode/png/exif-10                39164             30783 ns/op            4231 B/op        189 allocs/op
BenchmarkDecode/png/all-10                  5617            206111 ns/op           48611 B/op        310 allocs/op
BenchmarkDecode/webp/all-10                 3069            379637 ns/op          144181 B/op       2450 allocs/op
BenchmarkDecode/webp/xmp-10                 3291            359133 ns/op          139991 B/op       2265 allocs/op
BenchmarkDecode/webp/exif-10               47028             25788 ns/op            4255 B/op        191 allocs/op
BenchmarkDecode/jpg/exif-10                58701             20216 ns/op            4223 B/op        188 allocs/op
BenchmarkDecode/jpg/iptc-10               135777              8725 ns/op            1562 B/op         80 allocs/op
BenchmarkDecode/jpg/iptc/category-10      215674              5393 ns/op             456 B/op         15 allocs/op
BenchmarkDecode/jpg/iptc/city-10          192067              6201 ns/op             553 B/op         17 allocs/op
BenchmarkDecode/jpg/xmp-10                  3244            359436 ns/op          139861 B/op       2263 allocs/op
BenchmarkDecode/jpg/all-10                  2874            389489 ns/op          145700 B/op       2523 allocs/op
BenchmarkDecode/tiff/exif-10                2065            566786 ns/op          214089 B/op        282 allocs/op
BenchmarkDecode/tiff/iptc-10               16761             71003 ns/op            2603 B/op        133 allocs/op
BenchmarkDecode/tiff/all-10                 1267            933321 ns/op          356878 B/op       2668 allocs/op

When in doubt, Exiftool is right

The output of this library is tested against exiftool -n -json. This means, for example, that:

• We use f-numbers and not APEX for aperture values.
• We use seconds and not APEX for shutter speed values.
• EXIF field definitions are fetched from this table: https://exiftool.org/TagNames/EXIF.html
• IPTC field definitions are fetched from this table: https://exiftool.org/TagNames/IPTC.html
• The XMP handling is currently very simple, you can supply your own XMP handler (see the HandleXMP option) if you need more.

There are some subtle differences in output:

• Exiftool prints rationale number arrays as space formatted strings with a format/precision that seems unnecessary hard to replicate, so we use strconv.FormatFloat(f, 'f', -1, 64) for these.

Development

Many of the tests depends on generated golden files. To update these, run:

 go generate ./gen

Note that you need a working exiftool and updated binary in your PATH for this to work. This was tested OK with:

exiftool -ver
12.76

Debuggin tips:

 exiftool testdata/goexif_samples/has-lens-info.jpg -htmldump > dump.html