Metodebevisregistrering

DETR (Detection Transformer)

DETR (Detection Transformer) is an end-to-end framework for object detection introduced by Carion et al. in 2020 that reformulates detection as a direct set prediction problem using transformers. Unlike traditional approaches that use hand-crafted post-processing like non-maximum suppression, DETR treats object detection as a sequence-to-sequence problem where the transformer predicts all objects at once.

Sources recorded, not reviewed

Kilderegistrering

Citater kopieret ordret fra metodens kilderegistrering. Ingen påstandsniveauverifikation er udledt heraf.

End-to-End Object Detection with Transformers

Taksonomisk metoderegistrering · ml-model / deep-learning

Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., & Zagoruyko, S. (2020). End-to-end object detection with transformers. In European Conference on Computer Vision (pp. 213-229). Springer, Cham. · DOI 10.1007/978-3-030-58452-8_13

Åbn fuld metode

Kuraterede påstande

Påstande gemt i bevis-loggen, hver med sin egen vurdering.

Ingen kuraterede påstande endnu

Denne visning opfinder ikke en påstandsvurdering, når loggen ingen har.

Relaterede metoder

Genereret fra metodegrafen og vist som maskinelt foreslåede relationer — ingen bevispåstand er udledt.

Same method familyMasked Autoencodersmachine-suggested · Relational suggestion, not evidence.Same method familySegment Anything Modelmachine-suggested · Relational suggestion, not evidence.Same method familySwin Transformermachine-suggested · Relational suggestion, not evidence.Same method familyVision Mambamachine-suggested · Relational suggestion, not evidence.