Table 8.

Comparisons with other detectors on MS COCO test-dev. “MS${}_{train}$” denotes multi-scale training, otherwise using single-scale training. All experiments of our method set score_thr to 0.001, which slightly improves detection performance without a speed reduction. IoU-Aware R-CNN${}^{*}$ means that trainval dataset is used to train the detector and soft-NMS is employed at inference.

Method	Backbone	${\mathbf{MS}}_{\mathit{train}}$	AP	${\mathrm{AP}}_{50}$	${\mathrm{AP}}_{75}$	${\mathrm{AP}}_{\mathit{S}}$	${\mathrm{AP}}_{\mathit{M}}$	${\mathrm{AP}}_{\mathit{L}}$
one-stage detectors
SSD [12]	ResNet-101		31.2	50.4	33.3	10.2	34.5	49.8
RefineDet [30]	ResNet-101		36.4	57.5	39.5	16.6	39.9	51.4
RetinaNet [5]	ResNet-101		39.1	59.1	42.3	21.8	42.7	50.2
FSAF [17]	ResNet-101	✓	40.9	61.5	44.0	24.0	44.2	51.3
FSAF [17]	ResNeXt-101-64x4d	✓	42.9	63.8	46.3	26.6	46.2	52.7
FCOS [6]	ResNet-101	✓	41.5	60.7	45.0	24.4	44.8	51.6
FCOS [6]	ResNeXt-101-64x4d	✓	44.7	64.1	48.4	27.6	47.5	55.6
FoveaBox [31]	ResNet-101	✓	40.8	61.4	44.0	24.1	45.3	53.2
FoveaBox [31]	ResNeXt-101	✓	42.3	62.9	45.4	25.3	46.8	55.0
LTM [32]	ResNeXt-101-64x4d	✓	44.9	64.7	48.3	26.9	47.8	55.8
ATSS [33]	ResNeXt-101-32x8d	✓	45.1	63.9	49.1	27.9	48.2	54.6
two-stage detectors
Faster R-CNN [7]	ResNet-101		34.9	55.7	37.4	15.6	38.7	50.9
Faster R-CNN w/FPN [11]	ResNet-101		36.2	59.1	39.0	18.2	39.0	48.2
Mask R-CNN [34]	ResNeXt-101		39.8	62.3	43.4	22.1	43.2	51.2
Libra R-CNN [20]	ResNet-101		41.1	62.1	44.7	23.4	43.7	52.5
Libra R-CNN [20]	ResNeXt-101-64x4d		43.0	64.0	47.0	25.3	45.6	54.6
Grid R-CNN [35]	ResNet-101		41.5	60.9	44.5	23.3	44.9	53.1
Faster R-CNN w/ PISA [21]	ResNeXt-101		42.3	62.9	46.8	24.8	45.5	53.1
Cascade R-CNN [8]	ResNet-101		42.8	62.1	46.3	23.7	45.5	55.2
TridentNet [36]	ResNet-101	✓	42.7	63.6	46.5	23.9	46.6	56.6
IoU-Aware R-CNN	ResNet-50		40.7	59.8	44.0	22.9	43.5	51.2
IoU-Aware R-CNN	ResNet-101		42.3	61.3	45.7	23.3	45.5	54.5
IoU-Aware R-CNN	ResNeXt-101-32x4d		43.4	62.8	46.8	24.7	46.7	55.1
IoU-Aware R-CNN${}^{*}$	ResNeXt-101-32x4d		44.3	62.9	48.3	25.6	47.5	56.5