Require Pillow >= 9.4.0 to avoid AttributeError when loading image dataset

[albertvillanova]

Require Pillow >= 9.4.0 to avoid AttributeError when loading image dataset.

The PIL.Image.ExifTags that we use in our code was implemented in Pillow-9.4.0: python-pillow/Pillow@24a5405

The bug #6881 was introduced in datasets-2.19.0 by this PR:

• Transpose images with EXIF Orientation tag #6739

Fix #6881.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[albertvillanova]

Do you think this is worth making a patch release for?
CC: @huggingface/datasets

[github-actions]

Show benchmarks

PyArrow==8.0.0

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.005764 / 0.011353 (-0.005589)	0.004182 / 0.011008 (-0.006826)	0.064520 / 0.038508 (0.026012)	0.034260 / 0.023109 (0.011151)	0.245677 / 0.275898 (-0.030221)	0.277889 / 0.323480 (-0.045591)	0.004569 / 0.007986 (-0.003417)	0.002905 / 0.004328 (-0.001423)	0.049346 / 0.004250 (0.045095)	0.050529 / 0.037052 (0.013476)	0.264718 / 0.258489 (0.006229)	0.295705 / 0.293841 (0.001864)	0.028144 / 0.128546 (-0.100402)	0.011048 / 0.075646 (-0.064598)	0.206290 / 0.419271 (-0.212982)	0.035886 / 0.043533 (-0.007647)	0.245038 / 0.255139 (-0.010101)	0.269835 / 0.283200 (-0.013365)	0.018927 / 0.141683 (-0.122756)	1.136536 / 1.452155 (-0.315619)	1.183256 / 1.492716 (-0.309460)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.115372 / 0.018006 (0.097366)	0.315471 / 0.000490 (0.314982)	0.000238 / 0.000200 (0.000038)	0.000043 / 0.000054 (-0.000012)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.021201 / 0.037411 (-0.016210)	0.070374 / 0.014526 (0.055848)	0.077557 / 0.176557 (-0.099000)	0.124713 / 0.737135 (-0.612423)	0.078850 / 0.296338 (-0.217489)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.278674 / 0.215209 (0.063465)	2.739597 / 2.077655 (0.661942)	1.438214 / 1.504120 (-0.065906)	1.326373 / 1.541195 (-0.214822)	1.370961 / 1.468490 (-0.097529)	0.569160 / 4.584777 (-4.015617)	2.411890 / 3.745712 (-1.333822)	2.954073 / 5.269862 (-2.315788)	1.816883 / 4.565676 (-2.748794)	0.063123 / 0.424275 (-0.361152)	0.005531 / 0.007607 (-0.002076)	0.328184 / 0.226044 (0.102140)	3.263083 / 2.268929 (0.994155)	1.809159 / 55.444624 (-53.635465)	1.535257 / 6.876477 (-5.341220)	1.583428 / 2.142072 (-0.558644)	0.642950 / 4.805227 (-4.162277)	0.122240 / 6.500664 (-6.378424)	0.044596 / 0.075469 (-0.030873)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	0.999993 / 1.841788 (-0.841795)	12.941508 / 8.074308 (4.867200)	10.417519 / 10.191392 (0.226127)	0.134345 / 0.680424 (-0.546079)	0.014651 / 0.534201 (-0.519550)	0.288660 / 0.579283 (-0.290623)	0.274550 / 0.434364 (-0.159814)	0.327785 / 0.540337 (-0.212553)	0.422954 / 1.386936 (-0.963982)

PyArrow==latest

Show updated benchmarks!

Benchmark: benchmark_array_xd.json

metric	read_batch_formatted_as_numpy after write_array2d	read_batch_formatted_as_numpy after write_flattened_sequence	read_batch_formatted_as_numpy after write_nested_sequence	read_batch_unformated after write_array2d	read_batch_unformated after write_flattened_sequence	read_batch_unformated after write_nested_sequence	read_col_formatted_as_numpy after write_array2d	read_col_formatted_as_numpy after write_flattened_sequence	read_col_formatted_as_numpy after write_nested_sequence	read_col_unformated after write_array2d	read_col_unformated after write_flattened_sequence	read_col_unformated after write_nested_sequence	read_formatted_as_numpy after write_array2d	read_formatted_as_numpy after write_flattened_sequence	read_formatted_as_numpy after write_nested_sequence	read_unformated after write_array2d	read_unformated after write_flattened_sequence	read_unformated after write_nested_sequence	write_array2d	write_flattened_sequence	write_nested_sequence
new / old (diff)	0.006051 / 0.011353 (-0.005302)	0.003926 / 0.011008 (-0.007082)	0.051480 / 0.038508 (0.012972)	0.036102 / 0.023109 (0.012992)	0.273358 / 0.275898 (-0.002540)	0.293261 / 0.323480 (-0.030219)	0.004562 / 0.007986 (-0.003424)	0.002918 / 0.004328 (-0.001410)	0.050386 / 0.004250 (0.046135)	0.048427 / 0.037052 (0.011375)	0.280178 / 0.258489 (0.021689)	0.314599 / 0.293841 (0.020758)	0.030876 / 0.128546 (-0.097670)	0.010571 / 0.075646 (-0.065076)	0.058555 / 0.419271 (-0.360717)	0.034974 / 0.043533 (-0.008559)	0.266604 / 0.255139 (0.011465)	0.284712 / 0.283200 (0.001512)	0.020296 / 0.141683 (-0.121387)	1.116760 / 1.452155 (-0.335395)	1.157794 / 1.492716 (-0.334922)

Benchmark: benchmark_getitem_100B.json

metric	get_batch_of_1024_random_rows	get_batch_of_1024_rows	get_first_row	get_last_row
new / old (diff)	0.103777 / 0.018006 (0.085771)	0.314267 / 0.000490 (0.313778)	0.000226 / 0.000200 (0.000026)	0.000047 / 0.000054 (-0.000008)

Benchmark: benchmark_indices_mapping.json

metric	select	shard	shuffle	sort	train_test_split
new / old (diff)	0.023837 / 0.037411 (-0.013574)	0.082145 / 0.014526 (0.067619)	0.090434 / 0.176557 (-0.086123)	0.132096 / 0.737135 (-0.605040)	0.092426 / 0.296338 (-0.203913)

Benchmark: benchmark_iterating.json

metric	read 5000	read 50000	read_batch 50000 10	read_batch 50000 100	read_batch 50000 1000	read_formatted numpy 5000	read_formatted pandas 5000	read_formatted tensorflow 5000	read_formatted torch 5000	read_formatted_batch numpy 5000 10	read_formatted_batch numpy 5000 1000	shuffled read 5000	shuffled read 50000	shuffled read_batch 50000 10	shuffled read_batch 50000 100	shuffled read_batch 50000 1000	shuffled read_formatted numpy 5000	shuffled read_formatted_batch numpy 5000 10	shuffled read_formatted_batch numpy 5000 1000
new / old (diff)	0.299554 / 0.215209 (0.084345)	2.932382 / 2.077655 (0.854727)	1.549994 / 1.504120 (0.045874)	1.454944 / 1.541195 (-0.086251)	1.474987 / 1.468490 (0.006497)	0.586149 / 4.584777 (-3.998628)	0.972118 / 3.745712 (-2.773594)	2.991719 / 5.269862 (-2.278142)	1.876365 / 4.565676 (-2.689311)	0.065178 / 0.424275 (-0.359098)	0.005114 / 0.007607 (-0.002493)	0.353704 / 0.226044 (0.127660)	3.500940 / 2.268929 (1.232012)	1.965581 / 55.444624 (-53.479043)	1.662594 / 6.876477 (-5.213883)	1.702761 / 2.142072 (-0.439311)	0.663879 / 4.805227 (-4.141348)	0.120036 / 6.500664 (-6.380628)	0.043195 / 0.075469 (-0.032274)

Benchmark: benchmark_map_filter.json

metric	filter	map fast-tokenizer batched	map identity	map identity batched	map no-op batched	map no-op batched numpy	map no-op batched pandas	map no-op batched pytorch	map no-op batched tensorflow
new / old (diff)	0.997690 / 1.841788 (-0.844098)	13.448914 / 8.074308 (5.374606)	10.132469 / 10.191392 (-0.058923)	0.148493 / 0.680424 (-0.531930)	0.016670 / 0.534201 (-0.517531)	0.289708 / 0.579283 (-0.289575)	0.132938 / 0.434364 (-0.301425)	0.411425 / 0.540337 (-0.128913)	0.430748 / 1.386936 (-0.956188)

[lhoestq]

maybe not super important since it was not reported by users, this can be included in the next release

[Eric2i]

I observed the same AttributeError with Pillow == 10.3.0, while 9.4.0 works for me.

[lhoestq]

What's the error you're getting @Eric2i ?

On my side on 10.3.0 I could run this without errors:

import PIL.Image
PIL.Image.ExifTags.Base.Orientation is not None  # True

[Eric2i]

Sorry, false alarm. I double-checked that 10.3.0 is also good on my side. Thanks for your sample codes.