Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.
Introduced CVAT actions. Actions allow performing different predefined scenarios on annotations automatically (e.g. shape converters) (https://github.com/opencv/cvat/pull/7172)
The UI will now retry requests that were rejected due to rate limiting (https://github.com/opencv/cvat/pull/7216)
Update nvidia/cuda image version from 11.7.0 to 11.7.1 in transt serverless function. (https://github.com/opencv/cvat/pull/7124)
[Helm] Allow pre-release versions in kubernetes requirement to include AWS EKS versions (https://github.com/opencv/cvat/pull/7183)
GPU versions of serverless functions now use the latest-gpu Docker tag rather than latest (https://github.com/opencv/cvat/pull/7215)
[Compose, Helm] Downgraded KeyDB to 6.3.2 (https://github.com/opencv/cvat/pull/7236)
The GPU version of the YOLOv7 serverless function not actually using the GPU (https://github.com/opencv/cvat/pull/6940)
It is now possible to create Ground Truth jobs containing all frames in the task (https://github.com/opencv/cvat/pull/7126)
Incorrect Ground Truth chunks saving (https://github.com/opencv/cvat/pull/7126)
Reset source/target storage if related cloud storage has been deleted (https://github.com/opencv/cvat/pull/6801)
Prevent possible cyclic dependencies when enqueuing a rq job when ONE_RUNNING_JOB_IN_QUEUE_PER_USER is used (https://github.com/opencv/cvat/pull/7139)
Enqueue deferred jobs when their dependencies are moved to the failed job registry due to AbandonedJobError (https://github.com/opencv/cvat/pull/7139)
Reduce the number of requests to the server for task details (https://github.com/opencv/cvat/pull/7167)
Shape settings opacity and selected opacity reset on each frame change (https://github.com/opencv/cvat/pull/7186)
Server error in list quality settings API, when called in an org (https://github.com/opencv/cvat/pull/7190)
Incorrect handling of the hidden points in skeletons in quality comparisons (https://github.com/opencv/cvat/pull/7191)
[Helm] Fixed installing Traefik Middleware even if Traefik is disabled in the values (https://github.com/opencv/cvat/pull/7184)
Error code 500 when send change:frame event without duration. (https://github.com/opencv/cvat/pull/7211)
Added workaround for corrupted cached chunks (https://github.com/opencv/cvat/pull/7243, https://github.com/opencv/cvat/pull/7251)
This release has changes only in the Enterprise version.
CVAT now supports serverless Nuclio functions that return skeleton annotations. We've added a keypoint detector that supports skeletons for the following classes: body, head, foot, and hands. Deployment command: ./deploy_cpu.sh pytorch/mmpose/hrnet32/nuclio/ (https://github.com/opencv/cvat/pull/7033)
Implemented a feature that allows slicing one polygon/mask shape into two parts (https://github.com/opencv/cvat/pull/7084)
Implemented a feature that allows joining several masks into a single one (https://github.com/opencv/cvat/pull/7084)
[Helm] Introduced values that apply to all backend deployments/jobs (https://github.com/opencv/cvat/pull/7148)
The "use cache" option on the server is now ignored when creating a task with cloud storage data (https://github.com/opencv/cvat/pull/7087)
The Docker Compose file and Helm chart have been updated to enable Traefik access logs by default and change the log format to JSON (https://github.com/opencv/cvat/pull/7109)
[Helm] The PersistentVolumeClaim for the volume used to hold application data is now retained after uninstall (https://github.com/opencv/cvat/pull/7123)
[Helm] All backend-related deployments now use cvat-app as the value for the app label (https://github.com/opencv/cvat/pull/7127)
[Helm] The minimum compatible Kubernetes version is now 1.19.0 (https://github.com/opencv/cvat/pull/7132)
[Helm] The CVAT hostname can now be configured with ingress.hostname option (https://github.com/opencv/cvat/pull/7132)
[Helm] The ingress.tls configuration has been reworked. (https://github.com/opencv/cvat/pull/7132)
[Helm] The Traefik subchart updated to 25.0.0 (appVersion v2.10.5) (https://github.com/opencv/cvat/pull/7132)
[Docker Compose] Traefik updated to v2.10.* (https://github.com/opencv/cvat/pull/7150)
Support for V1 cloudstorages/id/content endpoint (https://github.com/opencv/cvat/pull/6946)
[Helm] ingress.hosts has been removed, use ingress.hostname instead. (https://github.com/opencv/cvat/pull/7132)
Fixed a data race condition during GT job creation (https://github.com/opencv/cvat/pull/7096)
Resolved an issue where the job state could not be changed multiple times without reloading the annotation view (https://github.com/opencv/cvat/pull/7158)
Corrected an issue where compressed chunks did not utilize the Exif rotation tag (https://github.com/opencv/cvat/pull/7162)
Minor styling issues on empty models page (https://github.com/opencv/cvat/pull/7164)
Fixed minor issue when brush marker is appended to a final mask (https://github.com/opencv/cvat/pull/7168)
Support for default bucket prefix (https://github.com/opencv/cvat/pull/6943)
Search for cloud storage and share files (https://github.com/opencv/cvat/pull/6943)
Ability to limit one user to one task at a time (https://github.com/opencv/cvat/pull/6975)
Support for using an external database in a Docker Compose-based deployment (https://github.com/opencv/cvat/pull/7055)
Migrated to rq 1.15.1 (https://github.com/opencv/cvat/pull/6975)
Compressed sequental change:frame events into one (https://github.com/opencv/cvat/pull/7048)
Create a local session for AWS S3 client instead of using the default global one (https://github.com/opencv/cvat/pull/7067)
Improved performance of chunk preparation when creating tasks (https://github.com/opencv/cvat/pull/7081)
Race condition in a task data upload request, which may lead to problems with task creation in some specific cases, such as multiple identical data requests at the same time (https://github.com/opencv/cvat/pull/7025)
Bug with viewing dependent RQ jobs for downloading resources from cloud storage when file path contains sub-directories. This is relevant for admins that can view detailed information about RQ queues. (https://github.com/opencv/cvat/pull/6975)
OpenCV.js memory leak with TrackerMIL (https://github.com/opencv/cvat/pull/7032)
Can't deploy detectron serverless function (https://github.com/opencv/cvat/pull/7047)
A mask becomes visible even if hidden after changing opacity level (https://github.com/opencv/cvat/pull/7060)
There is no switcher to personal workspace if an organization request failed (https://github.com/opencv/cvat/pull/7063)
A new feature allowing users to invite others to the organization via email. (https://github.com/opencv/cvat/pull/6901)
[SDK] In the SDK, a parameter has been introduced to TaskDataset which enables the option to disable annotation loading (https://github.com/opencv/cvat/pull/7019)
A test has been incorporated for retrieving bucket content in cases where the bucket includes manually created directories. (https://github.com/opencv/cvat/pull/7018)
The maximum length of the secret access key has been increased to 64 characters. (https://github.com/opencv/cvat/pull/6701)
The client will no longer load all organizations upon start (https://github.com/opencv/cvat/pull/7004)
The default value for Zookeeper from the Clickhouse subchart has been set to disabled. (https://github.com/opencv/cvat/pull/7003)
/api/projects, /api/tasks, and /api/jobs will no longer return information regarding the count of labels. This information was complicating SQL queries, making them hard to optimize. Instead, use /api/labels?task_id=tid or /api/labels?project_id=pid. (https://github.com/opencv/cvat/pull/6918)Issues causing potential double-sized file writes during task data uploading have been addressed. (https://github.com/opencv/cvat/pull/6952)
Issues encountered when retrieving CS content from GCS buckets containing manually created directories have been resolved. (https://github.com/opencv/cvat/pull/7006)
[SDK] In the SDK, cvat_sdk.auto_annotation.annotate_task has been optimized to avoid unnecessary fetching of existing annotations. (https://github.com/opencv/cvat/pull/7019)
The project/task/job update time is now correctly modified upon label updates. (https://github.com/opencv/cvat/pull/6958)
Enabled nginx proxy buffering (https://github.com/opencv/cvat/pull/6991)
Helm: set memory request for keydb (https://github.com/opencv/cvat/pull/6945)
Supervisord (https://github.com/opencv/cvat/pull/6945):
autorestart=true option for all workersPersist image filters across jobs (https://github.com/opencv/cvat/pull/6953)
Splitting skeleton tracks on jobs (https://github.com/opencv/cvat/pull/6968)
Uploading skeleton tracks in COCO Keypoints format (https://github.com/opencv/cvat/pull/6969)
Fixed Siammask tracker error on grayscale images (https://github.com/opencv/cvat/pull/6982)
Fixed memory leak on client side when event listener was not removed together with its context (https://github.com/opencv/cvat/pull/6984)
Fixed crash related to issue tries to mount to not existing parent (https://github.com/opencv/cvat/pull/6977)
Added 'notranslate' markers to avoid issues caused by extension translators (https://github.com/opencv/cvat/pull/6993)
Getting CS content when S3 bucket contains manually created directories (https://github.com/opencv/cvat/pull/6997)
Optimized huge memory consumption when working with masks in the interface (https://github.com/opencv/cvat/pull/6996)
Security upgrade opencv-python-headless from 4.5.5.62 to 4.8.1.78 (https://github.com/opencv/cvat/pull/6931)
Added X-Frame-Options: deny (https://github.com/opencv/cvat/pull/6992)
Full Changelog: https://github.com/opencv/cvat/compare/v2.7.5...v2.7.6
Full Changelog: https://github.com/opencv/cvat/compare/v2.7.4...v2.7.5
Full Changelog: https://github.com/opencv/cvat/compare/v2.7.3...v2.7.4
Full Changelog: https://github.com/opencv/cvat/compare/v2.7.2...v2.7.3