Computer Vision Datasets for AI Training
License image and video datasets from Wavebreak Media's owned archive, or commission new capture for your computer vision model. We produce the source media and can select content around your target classes, actions and environments.
Featured Datasets
Explore sample licensed datasets across image, video, audio, and text collections.
Computer Vision Training Data by Task
Choose the visual coverage your model needs. Annotation availability is confirmed for the selected collection or scoped separately.
Image Classification
Images or video frames selected around your class definitions, with variation in subjects, backgrounds and capture conditions.
Object Recognition and Detection
Target objects across viewpoints, scales and settings. Explore object recognition datasets for this task.
Image Segmentation
Source imagery for object or scene-region segmentation, with masks commissioned to your class definitions and boundary rules.
Human Activity and Motion
Video of actions, gestures and human-object interactions. Explore human activity recognition datasets.
Scene and Camera Motion
Environments, viewpoint changes and camera movement. See the Camera Motion Annotation Dataset for a focused collection.
Model Evaluation
Held-out images or video selected around deployment conditions and failure cases. Explore model evaluation datasets for dedicated testing projects.
Human-Centric Image and Video Coverage
Our archive covers people at work, at home and in everyday interactions. Select image datasets for frame-level tasks or video datasets when action over time matters.
Collections include first-person household activities and synchronized multi-view video. Where coverage is missing, custom production can target additional scenarios, participants or camera setups.
Frequently Asked Questions (FAQ)
Existing collections may include descriptions, categories and technical metadata. Project-specific work can include class labels, bounding boxes, masks or temporal segments. The specification defines the schema, annotation rules and acceptance criteria; these annotations are not included in every collection.
Agree grouping rules for related assets, such as the same subject, shoot or video sequence. Adjacent frames and near-duplicates should remain within one split. Source identifiers and available metadata support these checks; any limits on identifying related assets should be recorded.
Quotas can be set by class, environment or capture condition, subject to available coverage or production feasibility. Define background examples and hard negatives explicitly, including cases that resemble the target class but fall outside its inclusion rules.
Yes, when specified. Delivery can include source media and agreed derivatives, such as normalized images, transcoded clips or extracted frames. Identifiers preserve source relationships; frame timestamps and JSON, JSONL or CSV manifests can be included. See delivery and security options.
Start with your task, class coverage and deployment conditions. Volume also depends on visual variation and annotation detail. Evaluate a representative sample or pilot, then use observed model errors to identify where additional data is needed.
Commercial use is defined for the selected files in the licensing agreement. Available provenance and applicable model or property releases can be reviewed before purchase. See dataset licensing and compliance for the review process.
Request Computer Vision Datasets
Send your model task, target classes, approximate volume and annotation needs. We will assess existing collections, archive curation and new capture against your brief.
