Exactly what you receive

Assign one label per text item. Ideal for sentiment, topic tagging, or moderation.

• text string The original text content
• label string The assigned classification label
• agreement number Percentage of annotators who agreed
• annotators number Total annotators who labeled this item

Compare two responses and select which is better. Used for RLHF training data.

• text string Response A
• content_b string Response B
• label string Which is preferred
• agreement number Percentage of annotators who agreed

Assign multiple labels per item. For content tagging, medical coding, or topic detection.

• text string The original text
• labels array<string> All applicable labels for the item

Highlight and tag entities in text — people, places, organizations, dates.

• text string The original text
• entities array<object> List of tagged spans with start, end, text, label

Rate text on a 1–5 scale. Used for quality scoring, toxicity severity, or relevance ranking.

• text string The original text
• rating number Average rating across annotators
• ratings_all array<number> Individual ratings for QA

Assign one label per image. For medical imaging, product photos, or content moderation.

• image_url string URL to the image
• label string Classification label
• agreement number Inter-annotator agreement percentage

Compare two images and select the better one. For generative model training or ad creative.

• image_url string Image A
• image_url_b string Image B
• label string Preference choice

Draw rectangles around objects. The core of computer vision training.

• image_url string URL to the image
• boxes array<object> Bounding boxes with x, y, w, h, label

Draw free-form outlines around irregular shapes. For medical imaging, satellite, or autonomous driving.

• image_url string URL to the image
• polygons array<object> Polygons with label and list of {x,y} points

Place dots at anatomical landmarks. For sports biomechanics, fitness apps, or pose estimation.

• image_url string URL to the image
• keypoints array<object> Keypoints with x, y, label

Tag regions in documents — headers, tables, form fields, signatures. For invoice extraction and ID verification.

• document_url string URL to the document page
• regions array<object> Tagged regions with bounding box and label

Image + text + audio in one item. For VQA, image captioning, or medical image + note validation.

• image_url string Image URL (nullable)
• text string Contextual text or question
• audio_url string Audio URL (nullable)
• answer string Free-text annotator answer

Type what you hear. For call center QA, medical dictation, or African language audio.

• audio_url string URL to the audio file
• transcript string Full text of the audio
• word_count number Word count for payout calculation

Tag actions within video segments. For sports analytics, surgery, or robotic training.

• video_url string URL to the video
• segments array<object> Time segments with start, end, action
• duration number Total duration in seconds

Track objects across video frames. For autonomous vehicles, surveillance, or sports.

• video_url string URL to the video
• fps number Frames per second
• frames array<object> Bounding boxes per frame with frame number and label

Hand-object interaction labeling for robot training. The core of imitation learning.

• video_url string URL to first-person video
• hand_points array<object> Hand keypoints per frame
• object_boxes array<object> Object bounding boxes
• phases array<object> Task phases with start, end, action
• outcome string success or failure

3D bounding boxes in LiDAR/depth data. For autonomous vehicles, mining, and robotics.

• pcd_url string URL to the PCD or PLY file
• boxes_3d array<object> 3D boxes with x, y, z, w, h, d, label