Interesting, I’ve always wondered if the assignment problem could be solved as continuous optimization. I’m familiar with manifold optimization but didn’t realize the permutation matrix could be expressed as a manifold.
Has this been applied to networks like DETR for object detection? I’ve never been clear on how they made the Hungarian algorithm differentiable (and it seems like their approach has issues).
Has this been applied to networks like DETR for object detection? I’ve never been clear on how they made the Hungarian algorithm differentiable (and it seems like their approach has issues).