TOMB-AI

TOMB-AI (Tracking of Object Movement and Behavior) is an AI-powered software platform developed by Metofico for automated, marker-free animal tracking and behavior analysis. Designed for life sciences researchers, it eliminates the inefficiencies and inconsistencies of manual behavior scoring by automatically detecting, tracking, and quantifying animal behavior with high precision. The platform is fully no-code, making advanced behavioral analysis accessible to all researchers regardless of programming experience.

TOMB-AI operates through two distinct pipelines — Behavior Recognition and Multi-subject Tracking — each offering a comprehensive set of tools to streamline experimental data processing and exploration at scale.

Behavior Recognition Pipeline

• Automatic scoring of any behavior: The AI can accurately recognize any behavior that can be consistently scored by a human, covering a wide range of experimental paradigms including grooming, freezing, rearing, hanging, mounting, investigation, attack, and novel object interaction.
• Transparency and adaptability: TOMB-AI recognizes behaviors across any experimental setup by adapting to and learning from each condition using both spatial and temporal information, enabling reliable performance even in unseen footage.
• Flexible review tools: Extensive metrics enable easy data exploration, and users can review the AI's scoring, as well as add or delete behavioral events as needed.
• Distinguishing visually similar behaviors: The platform uses spatial and temporal context to differentiate between behaviors that appear similar, such as rearing versus eating, or rear versus half rear, providing more accurate recognition than pose-based approaches alone.
• Generalization across datasets: TOMB-AI demonstrates strong generalization to completely unseen footage, accurately recognizing behaviors across changes in color, lighting, mouse strain, and recording setup without requiring additional training.

Multi-subject Tracking Pipeline

• Markerless tracking of multiple subjects: TOMB-AI tracks multiple subjects simultaneously within the same setup without the need for physical markers, and is unaffected by subject color or setup variations.
• Compatible with any setup: Provides reliable tracking under any experimental condition, including both regular and infrared video, with the AI learning and adapting to each environment for optimal accuracy.
• Customizable tracking metrics: Automatically calculates key metrics such as heatmaps, track plots, speed, and distance traveled for each individual subject, streamlining data exploration.
• Flexible data visualization: Researchers can save track plots and heat maps at any time point during a video, giving full control over how data is explored and presented.
• Regions of Interest (ROI) analysis: An ROI feature enables deeper tracking insights by allowing analysis within defined spatial areas of the experimental setup.

Key Benefits

• Efficiency: TOMB-AI automates behavior scoring, significantly reducing manual effort and accelerating project completion compared to traditional methods.
• Reliability: Automated scoring eliminates human error and inconsistencies that arise from fatigue, attention span, and inter-rater variability.
• Repeatability: The platform ensures precise, repeatable analysis — the same data and parameters always produce the same results, regardless of the researcher's skill level.

Advantages Over Existing Solutions

• Unlike systems relying on pixel subtraction, TOMB-AI is significantly less affected by environmental variables such as lighting conditions, camera movement, or setup changes.
• Provides true behavior recognition rather than relying on proxy measures such as region entries, keypoint confirmations, or subject outline crossings, reducing false detections and improving reliability.
• Eliminates the years of development time and associated costs required to implement and extend open-source alternatives with behavior recognition capabilities.
• Outperforms competing tools such as MARS and human annotators in detecting social behaviors like attack and investigation by leveraging full spatial and temporal context.

Model Training and Data Requirements

• Training a behavior recognition model typically requires only 5 to 7 recordings, each lasting 2 to 10 minutes, featuring different subjects and containing examples of the behavior of interest.
• For models recognizing multiple behaviors simultaneously, a balanced representation of each behavior across the training videos is required.
• Metofico's team handles all model training — users simply share annotated videos of the behaviors of interest, and the team manages the rest.

Data Export and Output

• All analysis data can be downloaded as an Excel file for further processing.
• Individual metrics are available for download as PNG or SVG files, enabling customization of colors and styles for publication or reporting purposes.

TOMB-AI is continuously updated by Metofico based on customer feedback and advancements in the field. Free demos and trials are available, allowing research teams to evaluate how the platform can enhance their specific workflows before committing. For inquiries, Metofico can be contacted directly at [email protected].