Skip to content

Latest commit

 

History

History
173 lines (127 loc) · 10.2 KB

CHANGES.md

File metadata and controls

173 lines (127 loc) · 10.2 KB

CHANGES

Unreleased (latest)

  • Nothing new for the moment.

1.5.2 (2023-11-24)

  • Update pillow>=9.5.0 dependency to fix issued security vulnerability.

1.5.1 (2021-06-23)

  • Removed tkinter module incorrectly listed in requirement as package. It is preinstalled with Python.

1.5.0 (2021-06-14)

  • Add --srt/--vd-subtitles option that allows generation of an SRT subtitles-like file, but using Video Description (VD) metadata instead of traditional dialogue text. This metadata can be combined to most typical video players (eg: VLC) to rapidly display the video and annotated VD.

1.4.2 (2021-06-10)

  • Disable key sorting when generating a merged metadata file in YAML format to preserve insertion order.
  • Fix minor typos in documentation.

1.4.1 (2021-06-10)

  • Fix invalid alignment of different metadata sources that avoids pre-pending entries until the first occurrence start-time is reached against other sources. Start-time comparisons where incorrectly processed leading to missing metadata items. Issue was introduced in 1.3.0.

1.4.0 (2021-06-09)

  • Fix invalid links when using JSON reference generation ($ref notations) to ensure they can be properly loaded by packages such as jsonref.

  • Fix an issue where unspecified Video Inference (VI) metadata files would raise an error during argument parsing.

  • Support input files that could employ $ref links inside JSON or YAML metadata.

  • Support directly providing an input and pre-generated merged metadata file (--input-metadata) to be viewed with a video, skipping the metadata generation phase from the individual metadata file sources.

  • Add metadata total entries following merge strategy that can deduplicate original entries to align timestamps, and therefore do not necessarily match the original amount from the corresponding source. Both totals are reported.

1.3.0 (2021-06-09)

  • Provide the --text-auto and --text-manual options to define the Text Annotation method that was employed. If specified, merged result will contain the method: "<method>" entry or "undefined" value in the summary detail section of text_annotaitons metadata source.

  • Implement parsing and merging of Text Inference TSV files that provide action mappings between different types of lexical resources, embeddings generation and reference gold standard definitions against the original VD action. The new Text Inference metadata should be provided with --ti/--text-inference options similarly to other sources.

  • Update documentation about new Text Inference (TI) content format.

  • Avoid initial metadata entries of each source to be incorrectly added during merge or displayed in viewer until the corresponding first start time is reached.

1.2.0 (2021-06-07)

  • Add checkbox option to display only the central video-segment bounding boxes when provided from Video Inference metadata instead of enforced both central and dashed (outside delta central key frame) bounding boxes.
  • Fix invalid resolution of V1 Text Annotations against other version formats when the provided annotations were empty for some entries, and checked against to attempt detection of V4 precise definitions.
  • Fix handling of Text Annotations POS vs pos case and type vs iob according to parsed annotation version when displaying results on viewer.

1.1.0 (2021-05-20)

  • Support Text Annotations version 4:
    • Detect annotations that directly provide the same metadata format as V3 instead of annot_precises, and differentiate it from older format using annot_sentences (V2).

1.0.0 (2021-05-17)

  • Support Text Annotations version 3:

    • Detect and employ annot_precise metadata when provided in Text Annotations. This metadata contains similar information as annotations, but with additional fields that where not previously available (iob, offset, etc.).
    • Resolve modified field names (e.g.: POS becomes pos).
    • Handle new formats. Instead of a list of annotations object each with their POS, lemme, etc. fields, each of those fields are directly lists of same length matching the number of annotations.

  • Insert the current application version within the merged file to allow tracking. Merged files that do not provide version are generated with a prior version.

  • Add --references (-r) options to generate --merge file with references. Each time a VD, TA or VI item is used to generate the combined result, a UUID is placed instead of the literal object. All corresponding objects are then listed in separate mappings to avoid unnecessary duplication of data. When using this option, parsing of a file will require resolution of those references. Otherwise, the duplicated data is still returned (without -r flag).

  • Use explicit loading of JSON format when detected through json package since it parses large metadata much faster than yaml.

  • Remove unnecessary cary-over of annot and annot_precise original data from Text Annotations when generating merge, since the same information (different format) is already contained in the parsed and resolved merge result.

0.5.0 (2021-04-26)

  • Support Text Annotations version 2:
    • Directly employ provided sentence when available, or keep generating them otherwise.
    • Adjust dynamically the loaded annot_sentence used instead of annotations.

0.4.1 (2021-03-01)

  • Fix timestamp milliseconds parsing that caused errors in some cases because of too strict format.

0.4.0 (2021-01-26)

  • Support multiple video predictions provided through variable bounding boxes targets in the scene.
  • Display provided bounding boxes when available.

0.3.0 (2020-11-27)

  • Support multiple Video Inference metadata files simultaneously rendered in data fields.
  • Add better scrollbars support to allow moving around the data result fields.
  • Add initial merged file generation from other metadata sources.
  • Fix error of frame seek at the end of the video not able to properly retrieve the last metadata items.
  • Update the demo screenshot with the added metadata rendering.

0.2.0 (2020-11-17)

  • Add documentation.

    • Expected data formats and details.
    • Usage information to run the application.
    • Add demo screenshot of the displayed results.

  • Add snapshot button to save results of a single frame.

0.1.0 (2020-11-17)

  • First implementation of the result viewer.
    • Video display area.
    • Time slider to seek frames.
    • Buttons to control playback.
    • Text box to provide raw data.