Computer Vision News - August 2016

14 Computer Vision News Trick After the model has been trained (or loaded) it can be used to detect edges on unseen images. The edgesDetect function will do this. Some important parameters are: ● multiscale - run the structured edge detector on the original, half, and double resolution version of the image. ● sharpening - a procedure that aligns edge responses from overlapping Predictions. Intuitively, given a predicted segmentation mask, the mask can be morphed slightly so that it better matches the underlying image patch. Aligning overlapping masks to the underlying image data implicitly aligns the masks with each other, resulting in sharper, better localized edge responses. ● nTreesEval - set how many trees used for evaluation per location. ● NThreads - for speedup, number of threads for evaluation trees. ● nms - if true, apply non-maximum suppression to edges (i.e. set to zero elements that are not part of the given local maxima window). Here is the code snippet for the edgesDetect function; the function takes as input an image I and a structure with all the model parameters as explained above: And the result is: This is a significant improvement over the canny edge detector we have seen above. This time the grass area looks much more uniform and “clean”: it contains much fewer edges and the ones that remain have lower values. Still, some edges remain inside semantic regions (e.g. on the house walls). Trick