Skip to content

Latest commit

 

History

History
331 lines (176 loc) · 5.3 KB

HISTORY.md

File metadata and controls

331 lines (176 loc) · 5.3 KB
Raw

1.33.0

  • feat: support pyspark < 3 when distributing image-to-dataset job

1.32.0

  • feat: support min image size + max aspect ratio (@borisdayma)

1.31.0

  • feat: allow encoding in different formats (thanks @borisdayma)

1.30.2

  • Fix error message for incorrect input format

1.30.1

  • Bug fix: shard id was incorrect when resuming (thanks @lxj616)

1.30.0

  • Implement shard retrying

1.29.0

  • Validate input and output format
  • Implement incremental mode

1.28.0

  • use pyarrow in the reader to make it much faster

1.27.4

  • use 2022.1.0 of fsspec for python3.6

1.27.3

  • fix fsspec version

1.27.2

  • fix fsspec version

1.27.1

  • add gcsfs to pex

1.27.0

  • buffered writer fix: release ram more often
  • feat: accept numpy arrays (thanks @borisdayma)

1.26.0

  • add tfrecord output format (thanks @borisdayma)

1.25.6

  • fix an interaction between md5 and exif option

1.25.5

  • fix dependency ranges

1.25.4

  • use exifread-nocycle to avoid cycle in exifread

1.25.3

  • retry whole sharding if it fails

1.25.2

  • retry writing the shard in reader in case of error

1.25.1

  • small fix for logger and continuing
  • use time instead of perf_counter to measure shard duration

1.25.0

  • make metadata writer much faster by building the schema in the downloader instead of guessing it
  • add new option allowing to disable reencoding

1.24.1

  • hide opencv warning

1.24.0

  • force one thread for opencv
  • make total logger start time the minimum of workers start time
  • add s3fs into the released pex for convenience
  • make sharding faster on high latency fs by using a thread pool

1.23.1

  • fix logger on s3: do not use listing caching in logger

1.23.0

  • add tutorial on how to setup a spark cluster and use it for distributed img2dataset better aws s3 support:

  • initialize logger fs in subprocess to avoid moving fs over a fork()

  • use spawn instead of fork method

  • make total logging more intuitive and convenient by logging every worker return

1.22.3

  • fix release regex

1.22.2

  • fix fsspec support by using tmp_dir in main.py

1.22.1

  • fix pex creation

1.22.0

  • add option not to write

1.21.2

  • try catch in the logger for json.load
  • prevent error if logger sync is called when no call has been done
  • Add a build-pex target in Makefile and CI

1.21.1

  • decrease default log interval to 5s

1.21.0

  • add option to retry http download

1.20.2

  • add original_width by default for a consistent schema

1.20.1

  • fix relative path handling

1.20.0

  • Add multi distributor support : multiprocessing and pyspark

1.19.0

  • make the reader emits file paths instead of samples

1.18.0

  • use a logger process to make logging distribution friendly, also save json stat files next to folder/tar files

1.17.0

  • Use fsspec to support all filesystems

1.16.0

  • implement md5 of images feature

1.15.1

  • fix null convert in writer

1.15.0

  • add parquet writer

1.14.0

  • make reader memory efficient by using feather files

1.13.0

  • large refactoring of the whole code in submodules
  • Enhance image resize processing (esp re downscale) (@rwightman)

1.12.0

  • handle transparency (thanks @borisdayma)
  • add json input file support

1.11.0

  • Add support for .tsv.gz files (thanks @robvanvolt)

1.10.1

  • raise clean exception on image decoding error
  • remove the \n in urls for txt inputs
  • save the error message when resizing fails in metadata
  • add type hints to download function

1.10.0

  • use semaphores to decrease memory usage

1.9.9

  • fix an issue with resize_mode "no"

1.9.8

  • optimize listing files is back, sorted is eager so the iterator returned by iglob is ok

1.9.7

  • revert last commit, it could cause double iteration on an iterator which can cause surprising behaviors

1.9.6

  • optimize listing files (thanks @Skylion)

1.9.5

  • fix a bug affecting downloading multiple files

1.9.4

  • ensure sharded_images_to_dl is removed from memory at the end of downloading a file

1.9.3

  • solve the stemming issue: make keys uniques

1.9.2

  • Save empty caption if caption are none instead of not having the caption file

1.9.1

  • fix for the new logging feature when cleaning the status dict

1.9.0

  • wandb support is back

1.8.5

  • support for python 3.6

1.8.4

  • convert caption to str before writing

1.8.3

  • add back timeout properly

1.8.2

  • fixes

1.8.1

  • revert wandb for now, code is too complex and there are issues

1.8.0

  • feat: custom timeout (thanks @borisdayma)
  • feat: support wandb (thanks @borisdayma)

1.7.0

  • use albumentations for resizing (thanks @borisdayma)

1.6.1

  • depend on pyyaml to be able to use the last webdataset

1.6.0

  • feat: handle tsv + center crop (thanks @borisdayma)

1.5.3

  • increase stability by closing the pool and tarwriter explictely

1.5.2

  • improve memory usage

1.5.1

  • glob only input files of the right ext

1.5.0

  • add a save_additional_columns option

1.4.0

  • Multiple file support
  • Status dataframe

1.3.0

  • Uses a resizing method less prone to aliasing (thanks @skylion)
  • multi processing + multi threading

1.2.0

  • add webdataset support and benchmarks
  • supports reading as parquet and csv

1.1.1

  • fix cli

1.1.0

  • add image resizing mode

1.0.1

  • fixes

1.0.0

  • it works