Labelling Video Highlights for Food Videos

About OpenTrain

OpenTrain aggregates data-labeling and AI-training jobs from many AI companies and labeling platforms into a single job board so people can discover this kind of work without hunting dozens of sites.

Creating an OpenTrain account is free and applying takes only a few minutes — it’s the central place to find short-term and contractor opportunities in AI training.

About AI Training Work

AI training (also called data labeling, annotation, or human feedback) is the human side of building machine learning systems. For video tasks, human labelers mark precise moments, classify actions, and write visual descriptions that teach models how to see and understand video content.

This role focuses on food videos: your annotations will help models identify cooking steps, plating, tasting, and other food-specific events so AI can understand culinary content more accurately and safely.

The Role

You will identify and mark temporal segments in food videos that match natural-language queries, provide precise start/end boundaries, choose action/object labels, and write visually grounded descriptions for CLIP-style training. The work uses LABEL_STUDIO and is paid per labeled moment.

This is an entry-level, contractor position open worldwide. Compensation is PAY_PER_LABEL at USD 0.05 per label. Label types: ACTION_RECOGNITION and CLASSIFICATION. Data type: VIDEO.

• Employment type: Contractor
• Labeling software: Label Studio
• Payment: $0.05 per labeled moment (PAY_PER_LABEL, USD)
• Data type: Video; Label types: Action recognition and classification
• Open to applicants worldwide

What You'll Do (Task Workflow)

Follow the annotation guide for each query and produce precise, consistent labels. Tasks are performed in LABEL_STUDIO and require careful attention to visual detail and timing.

• Read the natural-language query and confirm the target action or event
• Watch the video and identify all segments that match the query (mark ALL matching segments)
• Use the timeline to mark start and end boundaries with high precision (target within 0.5 seconds)
• Classify the segment by selecting the action type and objects present
• Write a visual-proxy description that is action-focused and visually grounded for CLIP training
• Assign a confidence level (High, Medium, Low) based on boundary certainty

Domain Actions, Quality Guidelines & Common Mistakes

You will use a defined set of food-domain actions and must follow quality rules for boundary accuracy, specificity, and complete coverage.

• Common domain actions include: chopping ingredients, mise en place, mixing ingredients, kneading dough, sautéing, stirring the pot, deglazing, tasting food, adding seasoning, boiling, grilling, baking, plating the dish, garnishing, sauce drizzle, recipe introduction, finished dish reveal, eating rea
• Boundary accuracy: mark start/end within 0.5 seconds of the actual moment whenever possible
• Visual description guidelines: be visually grounded (e.g., "person in white chef coat slicing red tomatoes"), specific, action-focused, and CLIP-friendly
• Coverage: label every segment that matches the query, not just the first occurrence
• Avoid: marking segments that are too short or too long, vague descriptions, and abstract or subjective phrasing (e.g., avoid "delicious food")

Requirements & How to Apply

Required: a graduate degree plus demonstrable experience with data labelling and a strong understanding of food-video content. This listing is designated entry level but maintains the academic and domain requirement noted above.

You should be comfortable following detailed annotation guides, working in Label Studio, and applying precise temporal boundaries. Reliable internet access and the ability to work as a contractor are required. No specific languages or countries are required — this role is worldwide.

• Education: Graduate degree (required)
• Experience: Practical experience in data labelling (required) and familiarity with food videos
• Skill set: attention to detail, ability to produce time-accurate boundaries, clear visual descriptions, and basic familiarity with annotation tools (Label Studio)
• Work arrangement: contractor, remote (worldwide)
• How to apply: Create a free OpenTrain account and submit your application through the listing (applications are quick and take only a few minutes)