Related Experiment Video
Updated: Aug 6, 2026
26:43
Computer-Generated Animal Model Stimuli
Published on: July 29, 2007
Building models of animals from video
Deva Ramanan1, David A Forsyth, Kobus Barnard
1Toyota Technological Institute, Chicago, IL 60637, USA. ramanan@tti-c.org
Summary
This study presents an automated system for building 2D animal models from videos, enabling simultaneous tracking and recognition. The system enhances pictorial structures with novel texture descriptors for improved object detection and identification.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Object tracking, detection, and model building are often treated as distinct tasks.
- Existing methods may struggle with complex articulations and diverse image datasets.
Purpose of the Study:
- To demonstrate that tracking, object detection, and model building are unified activities.
- To develop an automated system for learning 2D articulated animal models from video.
- To enhance these models for robust animal recognition in diverse visual data.
Main Methods:
- A fully automatic system learns 2D pictorial structures from animal videos.
- Pictorial structures are augmented with a discriminative texture model and a novel texture descriptor.
- The system is evaluated on real video sequences and diverse image datasets (Corel, Google Web images).
Main Results:
- The system successfully tracks and identifies animals in video sequences.
- Learned models achieve good performance in recognizing animals from curated and web-scraped image sets.
- The approach demonstrates effective detection, localization, and part articulation recovery on challenging datasets.
Conclusions:
- Tracking, object detection, and model building can be integrated into a single framework.
- Augmenting pictorial structures with discriminative texture models significantly improves performance.
- The developed system offers a generalized approach to object recognition and tracking from video.
