Skip to content

Releases: cvat-ai/cvat

v2.4.4: General improvements in AI assistant

18 May 18:49
d5ef45a
Compare
Choose a tag to compare

Added

  • Introduced a new configuration option for controlling the invocation of Nuclio functions.
    (#6146)

Changed

  • Relocated SAM masks decoder to frontend operation.
    (#6019)
  • Switched person-reidentification-retail-0300 and faster_rcnn_inception_v2_coco Nuclio functions with person-reidentification-retail-0277 and faster_rcnn_inception_resnet_v2_atrous_coco respectively.
    (#6129)
  • Upgraded OpenVINO-based Nuclio functions to utilize the OpenVINO 2022.3 runtime.
    (#6129)

Fixed

  • Resolved issues with tracking multiple objects (30 and more) using the TransT tracker.
    (#6073)
  • Addressed azure.core.exceptions.ResourceExistsError: The specified blob already exists.
    (#6082)
  • Corrected image scaling issues when transitioning between images of different resolutions.
    (#6081)
  • Fixed inaccurate reporting of completed job counts.
    (#6098)
  • Allowed OpenVINO-based Nuclio functions to be deployed to Kubernetes.
    (#6129)
  • Improved skeleton size checks after drawing.
    (#6156)
  • Fixed HRNet CPU serverless function.
    (#6150)
  • Prevented sending of empty list of events.
    (#6154)

Full Changelog: v2.4.3...v2.4.4

v2.4.3: Small infrastructure improvements

25 Apr 15:34
8a2c23f
Compare
Choose a tag to compare

Changed

  • Docker images no longer include Ubuntu package sources or FFmpeg/OpenH264 sources
    (#6040)
  • TUS chunk size changed from 100 MB to 2 MB
    (#6058)

Full Changelog: v2.4.2...v2.4.3

v2.4.2: Segment Anything Model in CVAT

14 Apr 16:05
b4549f4
Compare
Choose a tag to compare

New Contributors

Full Changelog: v2.4.1...v2.4.2

v2.4.1: Optimized annotations fetching

05 Apr 18:40
a7f7ff1
Compare
Choose a tag to compare

Fixed

  • Optimized annotations fetching up to 10 times (#5974)
  • Incorrect calculation of working time in analytics (#5973)

v2.4.0: 3D Cuboids interpolation, advanced analytics, and PyTorch dataset adapter in SDK

16 Mar 08:36
61470b6
Compare
Choose a tag to compare

In this release, we’ve introduced new features based on our vision and suggestions in the CVAT community. Thank you for reaching out and providing us with valuable feedback. Here is the list of features we are shipping in the CVAT v2.4.0

  • Added the grid view and support of the contextual images for 2D and 3D tasks. For more information, see Contextual images.

    3D Cuboids interpolation

  • Added Interpolation for 3D cuboids. Also, annotation can be exported and imported in Datumaro and Supervisely Point Cloud formats.

  • Added Yolo v7 serverless functionally for automated annotation.

  • Collections in server API were redesigned for improved performance.

  • Added simple filters for collection endpoints in API. For example, now you can filter the tasks by username: /tasks?project_id=42&user=myusername

  • Improved analytics and enriched it with the following features: log collection from UI and server, exceptions collection, user activity visualization, aggregation of user and job working time, and log filtering for debugging purposes. Analytics are now based on ClickHouse + Vector + Grafana.

    Advanced analytics

  • Improved the work of the Pytorch adapter also known as the cvat_sdk.pytorch module, which allows you to use a CVAT project or task directly as a PyTorch-compatible dataset, eliminating the need for intermediate representation. For more information, see CVAT SDK Pytorch Adapter.

  • Improved high-level SDK APIs for task data uploading. The default data source is now the local file system. Added an option to wait for data processing to finish after data uploading, the option is turned on by default.

  • Added high-level API for working with organizations. For more information, see High-level API.

New Contributors

Full Changelog: v2.3.0...v2.4.0

v2.3.0: Brush tool, WebHooks, and Social auth

22 Dec 16:05
Compare
Choose a tag to compare

In this release, we’ve introduced new features based on our vision and suggestions in the CVAT community, plus addressed more than 20+ reported bugs. Thank you for reaching out and providing us with valuable feedback. Here is the list of features we are shipping:

  • Added the Brush tool to create pixel masks for disjoint objects with multiple parts: a house hiding behind trees, a car behind a pedestrian, or a pillar behind a traffic sign. Read more.

    brush

  • Added Webhooks for Projects and Organizations. Configure them to handle application notifications about changes in a specific project or organization. Read more.

    webhooks

  • Added sorting by Z-Order: farthest from the camera objects go on the top of the list. Now sorted objects have an additional indication on the objects sidebar.

    order

  • Enabled leveraging of NVIDIA GPU for YOLO v5 serverless function. Now you can annotate with YOLO faster.

  • Added social accounts authentication. Now you can log in to CVAT using your social auth accounts:

  • Added email authentication. Now you can authenticate in CVAT with your email instead of login.

    auth

  • Added Propagation backward to the Confirm Propagation widget, now propagation works both ways.

    propagate

New Contributors

Full Changelog: v2.2.0...v2.3.0
CHANGELOG.md: https://github.com/opencv/cvat/blob/v2.3.0/CHANGELOG.md

v2.2.0: Skeletons and SDK

12 Sep 20:30
Compare
Choose a tag to compare

Summary

What's Changed

New Contributors

Full Changelog: https://github.com/opencv/cvat/...

Read more

v2.1.0: Advanced filtration and sorting

09 Apr 02:55
3bd7c7e
Compare
Choose a tag to compare

Added

  • Task annotations importing via chunk uploads (#4327)
  • Advanced filtration and sorting for a list of tasks/projects/cloudstorages (#4403)
  • Project dataset importing via chunk uploads (#4485)
  • Support paginated list for job commits (#4482)

Changed

  • Added missing geos dependency into Dockerfile (#4451)
  • Improved helm chart readme (#4366)
  • Added helm chart support for CVAT 2.X and made ingress compatible with Kubernetes >=1.22 (#4448)

Fixed

  • Permission error occured when accessing the JobCommits (#4435)
  • job assignee can remove or update any issue created by the task owner (#4436)
  • Bug: Incorrect point deletion with keyboard shortcut (#4420)
  • some AI Tools were not sending responses properly (#4432)
  • Unable to upload annotations (#4513)
  • Fix build dependencies for Siammask (#4486)
  • Bug: Exif orientation information handled incorrectly (#4529)

New Contributors

v2.0.0: Advanced authorization and organizations

04 Mar 17:45
e33db5c
Compare
Choose a tag to compare

Summary

Added

  • Handle attributes coming from nuclio detectors (#3917)
  • Add additional environment variables for Nuclio configuration (#3894)
  • Add KITTI segmentation and detection format (#3757)
  • Add LFW format (#3770)
  • Add Cityscapes format (#3758)
  • Add Open Images V6 format (#3679)
  • Rotated bounding boxes (#3832)
  • Player option: Smooth image when zoom-in, enabled by default (#3933)
  • Google Cloud Storage support in UI (#3919)
  • Add project tasks pagination (#3910)
  • Add remove issue button (#3952)
  • Data sorting option (#3937)
  • Options to change font size & position of text labels on the canvas (#3972)
  • Add "tag" return type for automatic annotation in Nuclio (#3896)
  • Helm chart: Make user-data-permission-fix optional (#3994)
  • Advanced identity access management system, using open policy agent (#3788)
  • Organizations to create "shared space" for different groups of users (#3788)
  • Dataset importing to a project (#3790)
  • User is able to customize information that text labels show (#4029)
  • Support for uploading manifest with any name (#4041)
  • Added information about OpenVINO toolkit to login page (#4077)
  • Support for working with ellipses (#4062)
  • Add several flags to task creation CLI (#4119)
  • Add YOLOv5 serverless function for automatic annotation (#4178)
  • Add possibility to change git repository and git export format from already created task (#3886)
  • Basic page with jobs list, basic filtration to this list (#4258)
  • Added OpenCV.js TrackerMIL as tracking tool (#4200)
  • Ability to continue working from the latest frame where an annotator was before (#4297)
  • GET /api/jobs/<id>/commits was implemented (#4368)
  • Advanced filtration and sorting for a list of jobs (#4319)

Changed

  • Users don't have access to a task object anymore if they are assigned only on some jobs of the task (#3788)
  • Different resources (tasks, projects) are not visible anymore for all CVAT instance users by default (#3788)
  • API versioning scheme: using accept header versioning instead of namespace versioning (#4239)
  • Replaced 'django_sendfile' with 'django_sendfile2' (#4267)
  • Use drf-spectacular instead of drf-yasg for swagger documentation (#4210)

Deprecated

  • Job field "status" is not used in UI anymore, but it has not been removed from the database yet (#3788)

Removed

  • Review rating, reviewer field from the job instance (use assignee field together with stage field instead) (#3788)
  • Training django app (#4330)
  • v1 api version support (#4332)

Fixed

  • Fixed Interaction handler keyboard handlers (#3881)
  • Points of invisible shapes are visible in autobordering (#3931)
  • Order of the label attributes in the object item details(#3945)
  • Order of labels in tasks and projects (#3987)
  • Fixed task creating with large files via webpage (#3692)
  • Added information to export CVAT_HOST when performing local installation for accessing over network (#4014)
  • Fixed possible color collisions in the generated colormap (#4007)
  • Original pdf file is deleted when using share (#3967)
  • Order in an annotation file(#4087)
  • Fixed task data upload progressbar (#4134)
  • Email in org invitations is case sensitive (#4153)
  • Caching for tasks and jobs can lead to an exception if its assignee user is removed (#4165)
  • Added intelligent function when paste labels to another task (#4161)
  • Uncaught TypeError: this.el.node.getScreenCTM() is null in Firefox (#4175)
  • Bug: canvas is busy when start playing, start resizing a shape and do not release the mouse cursor (#4151)
  • Bug: could not receive frame N. TypeError: Cannot read properties of undefined (reding "filename") (#4187)
  • Cannot choose a dataset format for a linked repository if a task type is annotation (#4203)
  • Fixed tus upload error over https (#4154)
  • Issues disappear when rescale a browser (#4189)
  • Auth token key is not returned when registering without email verification (#4092)
  • Error in create project from backup for standard 3D ann...
Read more

v2.0.0-alpha

26 Dec 19:59
b7a50ac
Compare
Choose a tag to compare
v2.0.0-alpha Pre-release
Pre-release

Summary

  • Added OpenVINO logo and links to OpenVINO toolkit to the CVAT login page

    image

  • Implemented advanced identity access management system (IAM) using Open Policy Agent. Now CVAT has configurable rules to provide access to different resources on the server. Users can create organizations to manage resources together. We had more than 15 issues on GitHub which described different use cases from community and Intel customers (e.g., #3950, #2708, #2637, #2681).

    cvat_org

  • Implemented rotated bounding boxes. The feature was requested by our community multiple times to annotate objects in different scenarios.

    cvat_rbox

  • Added new annotation formats for import/export of datasets: KITTI, LFW, Cityscapes, Open Images v6

  • Implemented backup of projects to move them from one CVAT instance to another easily (previously it was implemented only for annotation tasks):

    cvat_backup

  • Implemented import of datasets. Now it is possible to create a project from datasets in MS COCO, Pascal VOC, ImageNet and other formats.

Details

Added

  • Add additional environment variables for Nuclio configuration (#3894)
  • Add KITTI segmentation and detection format (#3757)
  • Add LFW format (#3770)
  • Add Cityscapes format (#3758)
  • Add Open Images V6 format (#3679)
  • Rotated bounding boxes (#3832)
  • Player option: Smooth image when zoom-in, enabled by default (#3933)
  • Google Cloud Storage support in UI (#3919)
  • Add project tasks paginations (#3910)
  • Add remove issue button (#3952)
  • Data sorting option (#3937)
  • Options to change font size & position of text labels on the canvas (#3972)
  • Add "tag" return type for automatic annotation in Nuclio (#3896)
  • Advanced identity access management system, using open policy agent (#3788)
  • Organizations to create "shared space" for different groups of users (#3788)
  • Dataset importing to a project (#3790)
  • User is able to customize information that text labels show (#4029)
  • Support for uploading manifest with any name (#4041)
  • Added information about OpenVINO toolkit to login page (#4077)

Changed

  • Users don't have access to a task object anymore if they are assigneed only on some jobs of the task (#3788)
  • Different resources (tasks, projects) are not visible anymore for all CVAT instance users by default (#3788)

Deprecated

  • Job field "status" is not used in UI anymore, but it has not been removed from the database yet (#3788)

Removed

  • Review rating, reviewer field from the job instance (use assignee field together with stage field instead) (#3788)

Fixed

  • Fixed Interaction handler keyboard handlers (#3881)
  • Points of invisible shapes are visible in autobordering (#3931)
  • Order of the label attributes in the object item details(#3945)
  • Order of labels in tasks and projects (#3987)
  • Fixed task creating with large files via webpage (#3692)
  • Added information to export CVAT_HOST when performing local installation for accessing over network (#4014)
  • Fixed possible color collisions in the generated colormap (#4007)
  • Original pdf file is deleted when using share(#3967)

Security

  • Updated ELK to 6.8.22 which uses log4j 2.17.0 (#4052)