CVAT vs Labellerr

CVAT turns images, video, 3D data and audio into labeled datasets for AI workflows. Its annotation tools include boxes, masks, polygons, keypoints, skeletons and object tracks, with additional listed types such as cuboids, ellipses and tags. Teams can review jobs through validation and acceptance stages and assess quality using ground truth, honeypots and consensus workflows. Storage connections include Amazon S3, Azure Blob Storage, Google Cloud Storage, compatible buckets and self-hosted storage. CVAT names SAM 2, SAM 3, Ultralytics, Hugging Face and custom models as labeling aids. It exports to more than 20 formats, including COCO, YOLO, KITTI, Cityscapes and Pascal VOC, and offers API, SDK and CLI automation. Deployment can be managed in the cloud or self-hosted, including on-premises and air-gapped options. Free plans are available; online paid plans start at $23/user/mo billed annually. CVAT lists security controls but says it does not hold ISO 27001 or SOC 2 certifications.

Labellerr is a data labeling and annotation platform for AI and machine-learning teams preparing training data. Projects can use images, videos, PDFs, text, and audio. Annotation options include bounding boxes, polygons, polylines, keypoints, dots, classification, segmentation, and object detection. Automated labeling includes prompt-based and model-assisted methods, plus active learning; the company also states native support for Meta’s SAM, SAM 2, and SAM 3. A Python SDK can manage datasets and workflows in ML pipelines. Annotated data can be exported in formats such as COCO JSON, YOLO, Pascal VOC XML, CSV, and custom formats. Remote storage options include AWS S3, Google Cloud Storage, and Azure Blob; integrations with Vertex AI, SageMaker, and Azure ML are add-ons for applicable plans. Stated security measures include Auth0 authentication, TLS 1.2 or higher in transit, and AES-256 encryption. The free Researcher plan includes 2,500 data credits, one seat, one workspace, and up to 10 projects. Pro costs $9,999 a year, and unused data credits do not roll into the next payment cycle.