DarkLabel - 비디오/이미지 객체 레이블링 툴 (Video/Image Labeling and Annotation Tool)

동영상 또는 이미지에 ID를 가진 물체 경계사각형(bounding box)를 labeling할 수 있는 유틸러티 프로그램입니다. 인터페이스 등 사용성은 어느 프로그램보다 뛰어나다고 생각합니다. 학습 데이터 생성 외에도 비디오 편집이나 이미지들을 동영상으로 합치거나 분리할 때, 개인정보 보호를 위해 사람 얼굴 등을 모자이크하는 용도로도 활용할 수 있습니다.

 

※ 이 프로그램은 코드서명을 하지 않았기 때문에 Windows 및 웹브라우져에 의해 다운로드 및 실행이 차단될 수 있습니다 (참고링크). 100% 안전한 프로그램이긴 합니다만 필요한 분들만 사용하시기 바랍니다..

 

 

This is a utility program that can label object bounding boxes with ID in videos and images. Anyone can use it for non-commercial purposes. If you have any inquiry or suggestion, you can leave message below or email to darkpgmr.lee at gmail dot com.

 

※ Since this program is not code-signed, download and execution may be blocked by Windows and web browsers. This is a personally created program and a 100% safe program, but only for those who need it.

 

 

DarkLabel (Ver.2.4, 2021.09.04)

 

latest executable: DarkLabel2.4.zip

github: https://github.com/darkpgmr/DarkLabel

previous versions: check Version History below

 

DarkLabel2.0 Tutorial Video (https://www.youtube.com/watch?v=lok30aIZgUw)

 

old demo1(gt_extractor): https://www.youtube.com/watch?v=zFpX2bORz5E

old demo2(darklabel1.2): https://www.youtube.com/watch?v=vbydG78Al8s

 

 

Main Features (주요 기능 및 특징)

• Video/image object labeling & annotation tool (bounding box with ID and label)

- automatic object labeling by visual tracking (multi-target)

- semi-automatic labeling by linear interpolation

- user-configurable hotkeys and zoom in / zoom out support

- user-configurable data formats (pascal voc, darket yolo, xml/txt, any other user-defined formats)

• Video splitting (into images) & image merging (into a video file) tool
• Video cropping tool (cut and save only the selected section in the video)
• Video/image privacy masking tool (mosaic the box area in the image)
• Windows only (32/64 bits)

 

Program Configuration (프로그램 설정)

 

The program can be configured by modifying darklabel.yml attached in the zipped archive.

• define and modify data formats
• change hotkeys (frame navigation, action keys)
• change video/image export setting (video codec, frame rate, image format)
• set default save/load directory
• define class labels
• adjust GUI drawing (box width, box color, ...)

 

Basic Instruction (프로그램 사용법)

 

Arow/PgUp/PgDn/Home/End: navigate image frames

 

Mouse: Left(create box), Right(cancel the most recently created box)

Shift+Mouse: Left(modify box), Right(delete selected box/trajectory or all boxes)

Shift+DoubleClick: modify box properties (label, ID, difficulty)

 

DoubleClick: select/deselect box trajectory

*box trajectory: boxes connected across frames with the same ID and label

 

Ctrl+'+'/'-': zoom in/out

Ctrl+Arrow: scroll zoomed window

Ctrl+MouseWheel: zoom in/out

Ctrl+MouseDrag: scroll zoomed window

 

Enter or Spacebar: apply tracking (selected trajectories or newly created boxes only)

Ctrl+'s': save gt

F1: show this help

 

 

Advanced Instruction (고급 사용법)

 

• Labeling by object tracking

- By pressing 'Return' key, newly created bounding boxes in the current frame are tracked in the next frame and labeled automatically. (newly created boxes: the boxes that are created after entering the frame)

- If there are selected trajectories in the frame (object trajectory can be selected by double clicking a box), only the trajectory boxes are tracked (newly created boxes are not tracked).

- If tracking is applied to trajectories, it first deletes their tail parts (connected trajectory boxes after the current frame) and then continue the tracking, keeping their IDs and labels. This functionality can be used to correct previous wrong tracking.

- Two trackers are embeded. (developed by darkpgmr)
  . tracker1 (robust): good for deformable objects (e.g., human, ..)
  . tracker2 (accurate): good for rigid/static object (e.g., vehicle, wall, house, ...)

• Data Sampling/gathering

If you want to sample images in a video and save them (e.g. gathering training samples), draw dummy boxes on the images and then export the annotation results as images with the "no box drawing" option selected and the "labeled frames only" checked.

• Privacy Masking

draw boxes on the privacy area (e.g. human faces) and then export the annotation results with the "mosaic the box area" option selected.

• Full screen mode

do annotation in full screen mode in case the image is too large to show in your monitor (double click the title bar of image window). You also can utilize zoom in functionality.

• Don't forget to save your annotation results as often as possible. You can just press Ctrl+S.

 

Remark (2020.7.9)

 

그동안 프로그램이 이유없이 죽는다는 댓글들이 종종 있었는데, 원인을 모르다가 최근에야 UI delay를 없애기 위해 도입했던 thread 처리에 문제가 있었음을 알게 되었습니다. 이번 DarkLabel 2.0에서는 대부분  문제를 제거했지만 혹시 남은 문제가 발견되면 피드백 바랍니다.

 

Version History

 

* DarkLabel 2.4 (2021.09.04)

- support to save/load data in .csv file
- support MOT data format
- add a new format specifier ("c=n"): it can be used to save/load dummy values (see darklabel.yml for it's usage)
- enhance error handling (make it more robust to various external and internal errors)

 

* DarkLabel 2.3-update2 (2021.07.03)

- fix possible mismatch between labels and saved images in case of video input
- support to delete labels in intervals

- add two new adjustable parameters in darklabel.yml
  . min_object_size: minimum allowed box size
  . hotkey_return: secondary hotkey for return/enter key (for tracking)
- adjust GUI layout (make it more slim)

 

* DarkLabel 2.3-update1 (2021.03.23)

- fix the program to sort image list naturally by their file name

. before: 0.jpg, 1.jpg, 10.jpg, 100.jpg, 101.jpg, ...

. after: 0.jpg, 1.jpg, 2.jpg, 3.jpg, ...

 

* DarkLabel 2.3 (2020.10.25)

- enable selection of multiple object trajectories at the same time

- tracking is applied for object trajectories.

- minor bug fixes..

 

* DarkLabel 2.2 (2020.10.20)

- support media cropping (the selected section in the video / image list)

- it's able to change export settings on the GUI.

 

* DarkLabel 2.1 (2020.09.10)

- add two new parameters to support easy data gathering/sampling in darklabel.yml

  . export_labeled_frame_only

  . export_without_box_drawing

- doesn't popup message box now after video/image save is done.

 

* DarkLabel 2.0 (2020.06.29)

- support machine learning formats: xml(pascalvoc/imagenet), txt(darknet yolo)

- support user-defined data formats (data formats can be defined in 'darklabel.yml')

- correct program stability problem (bug-fix related to unexpected program exit)

- support zoom in / zoom out of working image

- add user-configurable hotkeys (it can be modified in 'darklabel.yml')

- labeling by tracking is now applied only to newly created boxes (keep the old boxes)

- lots of minor revision to enhance user convenience and program stability

 

* DarkLabel 1.3 (2018.03.31)

- 프로그램의 완성도를 높임 (ui, 반응속도, 안정성, 에러체크, 미세한 기능들, ...)

- trajectory 단위로 박스 선택/삭제/label수정 가능

- label(class)과 id를 분리해서 저장/로드할 수 있는 데이터 포맷 추가

 

* DarkLabel 1.2 (2018.02.19)

- 심각한 버그 수정: 숫자를 label로 사용할 경우 gt 데이터 로드시 오류 현상 수정

- interpolation을 이용한 labeling 기능 추가 (본문의 프로그램 사용법 참조)

- 개체 ID 자동부여 및 auto labeling 기능 추가 (본문의 프로그램 사용법 참조)

- 비디오내 프레임 이동시(navigation) GUI 반응성 개선

 

* DarkLabel 1.1 (2017.09.06)

- 작업결과를 동영상/이미지로 저장할 수 있는 기능 추가

 

* DarkLabel 1.0 (2017.07.18)

- labeling 기능 추가, tracker의 성능 및 UI 반응성 개선, 최신 비디오코덱 지원

- 프로그램 이름을 GT_Extractor에서 DarkLabel로 변경

 

* GT_Extractor 2.3 (2016.09.29)

- 버그 수정 (이미지 리스트에서 GT 파일 저장/불러올 때 오류 현상)

 

* GT_Extractor 2.2 (2016.08.26)

- 지원 비디오 포맷 추가 (avi, mpg, mp4, wmv, ...)

 

* GT_Extractor 2.1 (2013.08.12)

- 비디오 뿐만 아니라 이미지 목록에 대해서도 편집 가능

 

* GT_Extractor 2.0 (2013.01.15)

- 복수개의 대상 설정 및 추적 기능

 

* GT_Extractor 1.0 (2013.01.08)

- 이미지 당 하나의 대상만 설정할 수 있음

 

by 다크 프로그래머