Experimental results: here we show some experimental results of our model. The results was divided in 3 categories (results 1, 2 and 3) based on the segmentation quality results of each class (very good, good and acceptable, respectively), made visually by a specialist. The images shown a great variety of people, appearances, poses, image quality and thus, complexity. The dataset includes some images found in the web, as well some public images (used in other works) and images acquired in our lab.

Results 1: The images shown below illustrate some experimental results we consider very promising.
Please, click on the letter of the desired column to zoom it (A, B, C, D).


Results 2: The images shown below illustrate some experimental results we consider promising. Here we can see some minimal failures that could be improved in future works.
Please, click on the letter of the desired column to zoom it (A, B).


Results 3: Finally, the images shown below illustrate some experimental results we consider that could be improved. In such cases, the segmentation fail in some body parts (due to illumination problems, camouflage, occlusions, or maybe due to the simplicity of the proposed model, that can't handle all kind of images/poses/appearance, etc), but in general we can say it works fine.
Please, click on the letter of the desired column to zoom it (A, B).