ClassCut is a novel method for unsupervised class segmentation that alternates between segmenting object instances and learning a class model. It uses a segmentation energy defined over all images simultaneously, which can be efficiently optimized using techniques from interactive segmentation. The method progressively learns a class model by integrating observations across all images, capturing appearance, location, and shape of the class relative to an automatically determined coordinate frame common across images. This allows for stronger shape and location models, similar to those used in object class detection. ClassCut is fully automatic and learns class-specific models rather than object-specific ones. It outperforms GrabCut in unsupervised segmentation and offers competitive performance compared to state-of-the-art methods. The method is tested on the Caltech4, Caltech101, and Weizmann horses datasets, showing that it transfers class knowledge across images, improves segmentation results, and performs better than existing methods. The approach is inspired by interactive segmentation methods but is fully automatic. It uses a binary pairwise energy function and incorporates priors tailored for segmenting classes, including within-image smoothness, between-image smoothness, border penalty, and area reward. The class model includes appearance, location, and shape models, with the location and shape models relative to a reference frame determined automatically. The method is evaluated on three datasets, showing improved segmentation accuracy compared to existing methods. ClassCut outperforms GrabCut and spatial topic models, demonstrating its effectiveness in unsupervised class segmentation.ClassCut is a novel method for unsupervised class segmentation that alternates between segmenting object instances and learning a class model. It uses a segmentation energy defined over all images simultaneously, which can be efficiently optimized using techniques from interactive segmentation. The method progressively learns a class model by integrating observations across all images, capturing appearance, location, and shape of the class relative to an automatically determined coordinate frame common across images. This allows for stronger shape and location models, similar to those used in object class detection. ClassCut is fully automatic and learns class-specific models rather than object-specific ones. It outperforms GrabCut in unsupervised segmentation and offers competitive performance compared to state-of-the-art methods. The method is tested on the Caltech4, Caltech101, and Weizmann horses datasets, showing that it transfers class knowledge across images, improves segmentation results, and performs better than existing methods. The approach is inspired by interactive segmentation methods but is fully automatic. It uses a binary pairwise energy function and incorporates priors tailored for segmenting classes, including within-image smoothness, between-image smoothness, border penalty, and area reward. The class model includes appearance, location, and shape models, with the location and shape models relative to a reference frame determined automatically. The method is evaluated on three datasets, showing improved segmentation accuracy compared to existing methods. ClassCut outperforms GrabCut and spatial topic models, demonstrating its effectiveness in unsupervised class segmentation.