---
title: "End-to-End Object Detection with Transformers"
description: "DETR reframed object detection as direct set prediction: learned query slots propose boxes in parallel, Hungarian matching assigns one positive owner to each target, and unmatched slots learn no object. Follow the mechanism, the decoder ablations, and the scale-dependent COCO evidence."
canonical_url: "https://fanout.sh/daily/2026-10-31-end-to-end-object-detection-transformers"
md_url: "https://fanout.sh/daily/2026-10-31-end-to-end-object-detection-transformers.md"
last_updated: "2026-10-31"
access: "public"
---

# End-to-End Object Detection with Transformers

DETR reframed object detection as direct set prediction: learned query slots propose boxes in parallel, Hungarian matching assigns one positive owner to each target, and unmatched slots learn no object. Follow the mechanism, the decoder ablations, and the scale-dependent COCO evidence.

## Paper details

- Authors: Nicolas Carion et al.

- Venue: ECCV 2020

- Track: AI Research

- Difficulty: Hard

- Reading time: 13 min

- Original paper: https://arxiv.org/abs/2005.12872v3

## What you will learn

- How a fixed bank of learned object queries represents a variable-sized set through the no-object class.

- Why one-to-one Hungarian matching turns duplicate predictions into training errors instead of an inference-time cleanup problem.

- What the 42.0 AP headline hides about large and small objects, long training, and fixed query capacity.

## Continue exploring

[Compare grid-based YOLO](https://fanout.sh/daily/2026-09-10-yolo-unified-object-detection): Contrast DETR's learned set assignment with an earlier one-stage detector whose prediction responsibilities are tied to a spatial grid.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md