I've recently come across an amazing work for human pose estimation: DensePose: Dense Human Pose Estimation In The Wild by Facebook.

In this work, they have tackled the task of dense human pose estimation using discriminative trained models.

I do understand that "correspondence" means how well pixels in one image correspond to pixels in the second image (specifically, here - 2D to 3D).

But what does "dense" means in this case?


In computer vision, the adjectives dense and sparse are used in a variety of tasks (e.g. optimal flow), but they are commonly used in the context of the correspondence problem, which is the problem of finding a map (or correspondence) between pixels of two images (e.g. two successive frames of a video). In this context, these adjectives thus refer to the number of pixels of the image that are used to solve this specific task. A dense correspondence is thus a correspondence between two images using all (or, at least, many) pixels. In other words, a dense correspondence attempts to map all (or many) pixels of an image to all (or many) pixels of another image.

In the case of DensePose, the correspondences are between RGB images and surface-based representation of the human body (the figures of the paper illustrates this).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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