A Paper Presentation for "Structured Forests for Fast Edge Detection" by Dollár, Piotr, and C. Lawrence Zitnick at Computer Vision (ICCV), 2013 IEEE International Conference on. IEEE, 2013.
COMPUTING ANTI-DERIVATIVES(Integration by SUBSTITUTION)
Structured Forests for Fast Edge Detection [Paper Presentation]
1.
2. Dollár, Piotr, and C. Lawrence Zitnick. "Structured forests for fast edge detection.“ Computer Vision (ICCV), 2013 IEEE International Conference on. IEEE, 2013.
3. MainContribution
Compute edge maps in realtime,
faster than the competing state-of-the-art
Proposed
Method
Structured Random Forests
This presentation is inspired by the talk: http://techtalks.tv/talks/structured-forest-for-fast-edge-detection/59412/
33. Training Model
Just one difference to random forests:
clusterthe output into a binary or multiclass output using distance function
34. Clustering
푌: Structured space where information gain not well defined
퐶: Discrete space where information space is good defined
푍: Intermediate space where similarity measurement is easy to compute
Π∶푌→푍, 푍→퐶
35. Training Model
•Computing information gain
–Labels 퐶are discrete, standard entropy criterions used.
•Combining predictions
–To combine 푦1…푦푛∈푌into a prediction:
•Compute푧푖=Π휑(푦푖)of dimension 푚
•Select 푦푘, whose 푧푘=푎푟푔푚푖푛푧푘 푖,푗(푧푘푗−푧푖푗)2(medoid)
+ Computing medoids is fast, 푂(푛푚)
45. Results
•BSDS 500 image set
–Multi-scale ties or outperforms the accuracy of the state of the art.
–Single-scale improves runtime by 5x to 10x
46. Results
•BSDS 500 image set
–Multi-scale ties or outperforms the accuracy of the state of the art.
–Single-scale improves runtime by 5x to 10x
47. Results
•BSDS 500 image set
–Multi-scale ties or outperforms the accuracy of the state of the art.
–Single-scale improves runtime by 5x to 10x
48. Results
•NYU image set
–Multi-scale is slightly better than the state of the art.
–Improved performance by multiple orders of magnitude
49. Conclusions
•Realtimestructured learning method for edge detection
•General purpose method for learning structured random forests
•Real time + state of the art accuracy → new applications possible
•Novel learning approach may be applicable to other problems.