1 / 29

Sketch Tokens: A Learned Mid-level Representation for Contour and Object Detection

Sketch Tokens: A Learned Mid-level Representation for Contour and Object Detection. CVPR2013 Poster. Outline. Introduction Method Results Discussion. Introduction.

celina
Download Presentation

Sketch Tokens: A Learned Mid-level Representation for Contour and Object Detection

An Image/Link below is provided (as is) to download presentation 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. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Sketch Tokens: A Learned Mid-levelRepresentation for Contour and Object Detection CVPR2013 Poster

  2. Outline Introduction Method Results Discussion

  3. Introduction Figure 1. Examples of sketch tokens learned from hand drawn sketches represented using their mean contour structure. Notice the variety and richness of the sketch tokens.

  4. Introduction Wetypically utilize a few hundred tokens, which captures a majorityof the commonly occurring edge structures. an efficient approach that can compute per-pixeltoken labelings in about one second per image. We propose a novel approach to both learning and detecting local edge-based mid-level features.

  5. Method • Defining sketch token classes • Detecting sketch tokens

  6. Defining sketch token classes These include straight lines, t-junctions, y-junctions, corners, curves, parallel lines, etc.

  7. Detecting sketch tokens Feature extraction Classification

  8. Feature extraction Two types of features are then employed: features directly indexing into the channels self-similarity features

  9. features directly indexing into the channels： • channels are composed of color, gradient, and orientedgradient information in a patch extracted from acolor image. • Pixels in the resulting channelsserve as the first type of feature for our classifier.

  10. self-similarity features： • The self-similarity features capture the portionsof an image patch that contain similar textures basedon color or gradient information.

  11. For channel and grid cells and , we define the self-similarity feature as: where is the sum of grid cell in channel . An illustration of self-similarity features is shown in Fig. 3.

  12. Classification Two considerations must be taken into account when choosing a classifier for labeling sketch tokens in image patches. First, every pixel in the image must be labeled, so the classifiermust be efficient. Second, the number of potentialclasses for each patch ranges in the hundreds.

  13. Results Contour detection Object detection

  14. Contour detection If is the probabilityof patch belonging to token , and is the probabilityof belonging to the “nocontour” class, the estimated probabilityof the patch’s center containing a contour is:

  15. Contour detection results We test our contour detector on the popular BerkeleySegmentation Dataset and Benchmark (BSDS500).

  16. Object detection INRIA pedestrian PASCAL VOC 2007

  17. INRIA pedestrian： For pedestrian detection we use an improvedimplementationof Doll´ar et al. that utilizes multiple image channelsas features for a boosted detector .

  18. PASCAL VOC 2007： Our final set of results use the PASCAL VOC 2007dataset. The dataset contains real world images with 20labeled object categories such as people, dogs,chairs, etc.

  19. Discussion Discovering new sets of observable mid-levelinformation that may be used for feature learning is an interestingand open question. We’ve explored several in this paper, butother tasks may also benefit.

More Related