End-to-End Object Detection with Transformers

26 May 2020Nicolas CarionFrancisco MassaGabriel SynnaeveNicolas UsunierAlexander KirillovSergey Zagoruyko

We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression procedure or anchor generation that explicitly encode our prior knowledge about the task... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Object Detection COCO minival Faster RCNN-R101-FPN+ box AP 44 # 23
AP50 63.9 # 14
AP75 47.8 # 13
APS 27.2 # 11
APM 48.1 # 12
APL 56 # 22
Object Detection COCO minival DETR-DC5 (ResNet-101) box AP 44.9 # 18
AP50 64.7 # 12
AP75 47.7 # 14
APS 23.7 # 25
APM 49.5 # 8
APL 62.3 # 6
Panoptic Segmentation COCO panoptic PanopticFPN++ PQ 44.1 # 3
SQ 79.5 # 2
RQ 53.3 # 2
PQth 51.0 # 1
SQth 83.2 # 1
RQth 60.6 # 2
PQst 33.6 # 2
SQst 74.0 # 2
RQst 42.1 # 2
AP 39.7 # 1
Panoptic Segmentation COCO panoptic DETR-R101 (ResNet-101) PQ 45.1 # 2
SQ 79.9 # 1
RQ 55.5 # 1
PQth 50.5 # 2
SQth 80.9 # 2
RQth 61.7 # 1
PQst 37 # 1
SQst 78.5 # 1
RQst 46 # 1
AP 33 # 2

Methods used in the Paper