I wonder if there is a method works as follows: 1) The algorithm is trained with some images of a particular object (I train my Image with 1000 banana images) 2) I use the training data to segment a test image (semantic segmentation)

The critical point here is, the training images DO NOT contain labeling information (foreground - background). I expect the algorithm to learn the shape of the object with some common edge orientations or features etc.

I study over Graphical Models for Segmentation and the most similar method is "OBJCUT" proposed by the following paper:


But, they train the system with shape exemplars (ground truth images contain segmentations done by humans).

My objects to segment are non-articulated objects.


1 Answer 1


This is not a complete answer (I would comment directly your question but not allowed to due to rep), but you could look at "Learning and detecting shape models v1.3 (September 2009)" (http://groups.inf.ed.ac.uk/calvin/software.html). The paper's name is: "From images to shape models for object detection" by Ferrari, Jurie & Schmid

The idea is to learn a contour based model of object which is then possible to deform. Actually, you have to segment a little bit by providing simply bounding boxes (x-y coordinates of the top-left and bottom-right corner) of the object in the training images. You may also use training images that are already cropped around the object which you want to learn the model.

I tried their Matlab implementation and creating the bounding boxes for my training images was about as fast as manually cropping them in Gimp or else...

EDIT: BTW, if you wonder how to create groundtruth files with this program, I used a text editor such as "Notepad++" to create a text file. In the file, I wrote, separated by spaces, the coordinates of the bounding box, e.g.:" 10 33 572 458 " (no quote) then saved the file with an appropriate name (check how they do for their sample files) with the dummy extension ".groundtruth", their Matlab implementation will recognize the files.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.