I have near about 70-75 images with segmentation ground-truth. My problem is 3 class problem. Super pixel classification using feature classification is done by other implementation. But for better segmentation smoothing, I want to apply CRF using unsupervised learning (without using any other models except superpixel classifier model) from the segmented output image. I have two queries:
(1) Already I have computed the superpixel classification result? In that .mat file the result is kept in a 3D matrix (image_row_image_col_No_of_Classes). Is this .mat array is compatible with the intended classProbs object to be inputted in uflow.inferenceSuperPixel() function if not what is the structure of the .mat file classProbs object?
(2) How adjacency probs is computed. What is the structure of the .mat file for prob? If the adjacency probs is kept none is there any problem?
I have near about 70-75 images with segmentation ground-truth. My problem is 3 class problem. Super pixel classification using feature classification is done by other implementation. But for better segmentation smoothing, I want to apply CRF using unsupervised learning (without using any other models except superpixel classifier model) from the segmented output image. I have two queries:
(1) Already I have computed the superpixel classification result? In that .mat file the result is kept in a 3D matrix (image_row_image_col_No_of_Classes). Is this .mat array is compatible with the intended classProbs object to be inputted in uflow.inferenceSuperPixel() function if not what is the structure of the .mat file classProbs object?
(2) How adjacency probs is computed. What is the structure of the .mat file for prob? If the adjacency probs is kept none is there any problem?