Transformers for detection

Sparse R-CNN

Fully sparse detection approach that uses a fixed set of learnable proposed boxes and a cascade of transformers to refine predictions, eliminating the need for heuristics like anchors or NMS.

← Indietro