Detection by Detections: Non-parametric Detector Adaptation for a Video. Outline. Introduction Non-parametric detector adaption Binary codes with a vocabulary tree Similarity measure of the binary codes Transfer classification Identity grouping of detections Experiment Conclusion.
Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.
Detection by Detections: Non-parametric Detector Adaptation for a Video
Extend any static-image-based object detector to video object detection.
Needs neither the original training data, nor manually labeled online examples.
1.Set generic human detector work on high recall low precision point
2.To build a vocabulary tree using hierarchical k-means
3.Encode the set of most confident visual detections as a set of binary vectors
 D. Nister and H. Stewenius. Scalable recognition with a vocabulary tree. In CVPR, 2006.
Define a mapping T
||c||: the number of non-zero bits in the binary vector c
compose positive pool
Group ID g(u) of any other example u
The normalization scheme evaluation
The evaluation on the threshold for positive pool
The tree depth exploration
Performance of video object detection
Data is from the EC Funded CAVIAR project/IST 2001 37540
Present the adaption performance based on 
 P. Felzenszwalb, D. McAllester, and D. Ramanan. A discriminatively
trained, multiscale, deformable part model. In
F(mode(gi)) counts the number of examples in group i with identity mode(gi)
On positive pool
On all detection
Difficult to be detected objects in video
First row by our approach
Second row by k-means clustering
Simple and effective solution to improve the pure detection accuracy of off-shelf detectors trained from static images on target videos.