Toward Learning Mixture-of-Parts Pictorial Structures
Lecture slides:
- Talk Objectives
- The OSU Digital Scout Project
- Raw Video Data
- Registered Video Data
- Problem: Formation Labelling
- Challenges in Formation Labelling
- Problem Constraints
- Previous Attempt
- Structured Output Representations
- Pictorial Structures
- Matching pictorial structures
- Efficient matching
- Pictorial Structures for Football
- Mixture of Parts Pictorial Structures (MoPPS)
- MoPPS Inference
- MoPPS Tree Representation
- MoPPS Tree for Football
- MoPPS Tree Inference
- Branch-and-Bound Search
- Lower Bound Computations
- Upper Bound Computations
- MoPPS Tree Parameters for Football
- Background Model
- Histogram-Based likelihood
- Experiments
- Results
- Anytime Behavior: % Correct
- Directions Learning MoPPS Models
- Structure and Parameter Learning
- Speedup Learning
- Active Model Calibration
- Summary and Future Work
- The OSU
Author: Alan Fern, Oregon State University