Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, & Sergey Zagoruyko (2020)
Lecture Notes in Computer Science, 213-229.
DOI: https://doi.org/10.1007/978-3-030-58452-8_13
Abstract. Introduces DETR, which reframes object detection as a set prediction problem using a transformer encoder-decoder and bipartite matching loss, eliminating hand-crafted components such as anchors and non-maximum suppression.
Tags: object-detection transformers detr