30 likes | 119 Views
This project focuses on leveraging temporal connections between concepts to enhance concept detection performance over time periods. By considering the causality between concepts, the aim is to improve the overall concept detection accuracy using TRECVID and soccer datasets.
E N D
Improving Concept Detection by Utilizing Temporal Relationships REU: Levi Smith Mentor: KhurramSoomro
Overview • We aim to use the temporal relationships between concepts, to improve concept detection over a temporal period • Rather than looking at a single concept in isolation, we want to take advantage of what we define as the causality between two concepts • We want to improve concept detection in general and have applied our methods to two datasets: TRECVID and a self-annotated soccer dataset Improving Concept Detection by Utilizing Temporal Relationships – Levi Smith t t
Causality • In complex events, these temporal relationships exist and can be easily seen in the case of soccer • There is often confusion in the classifier probabilities, which we want to correct • We can use this causal information to give more confidence to the most probable concept that will occur next in a sequence Attempt Corner No Concept Celebration Goal 12 Second Interval Improving Concept Detection by Utilizing Temporal Relationships – Levi Smith