SimpleDet Model Zoo

Introduction

This file documents a large collection of baselines trained with SimpleDet.

Common Settings

• All models were trained on train2014+valminusminival2014, and tested on minival2014.
• We adopt the same training schedules as Detectron. 1x indicates 6 epochs and 2x indicates 12 epochs since we append flipped images into training data.
• We report the training GPU memory as what nvidia-smi shows.

ImageNet Pretrained Models

We provide the ImageNet pretrained models used by SimpleDet. Unless otherwise noted, these models are trained on the standard ImageNet-1k dataset.

• resnet-v1-50: converted copy of MSRA's original ResNet-50 model.
• resnet-v1-101: converted copy of MSRA's original ResNet-101 model.
• resnet-50: ResNet-v2-50 model provided by MXNet Model Gallery.
• resnet-101: ResNet-v2-101 model provided by MXNet Model Gallery.
• resnet50_v1b: converted from gluoncv.
• resnet101_v1b: converted from gluoncv.
• resnet152_v1b: converted from gluoncv.
• resnext-101-64x4d: converted copy of FB's original ResNeXt-101-64x4d model.
• resnext-101-32x8d: converted copy of FB's ResNeXt-101-32x8d model.
• resnext-152-32x8d-IN5k: converted copy of FB's ResNeXt-152-32x8d-IN5k model trained on ImageNet-5k.

ResNetV1b Baselines

All config files can be found in config/resnet_v1b. Pretrains are converted from GluonCV. All AP results are reported on minival2014 of the COCO dataset.

Model	Backbone	Head	Train Schedule	AP	AP50	AP75	APs	APm	APl
Faster	R50v1b-C4	C5-512ROI	1X	35.7	56.7	37.9	18.6	40.4	48.1
Faster	R50v1b-C4	C5-512ROI	2X	36.9	57.9	39.3	19.9	41.4	50.2
Faster	R101v1b-C4	C5-512ROI	1X	40.0	61.3	43.1	21.5	44.8	54.3
Faster	R101v1b-C4	C5-512ROI	2X	40.5	61.2	43.8	22.5	44.8	55.4
Faster	R152v1b-C4	C5-512ROI	1X	41.3	62.6	44.6	23.4	46.2	55.6
Faster	R152v1b-C4	C5-512ROI	2X	41.8	62.4	45.2	23.2	46.0	56.9
Faster	R50v1b-FPN	2MLP	1X	37.2	59.4	40.4	22.3	41.3	47.6
Faster	R50v1b-FPN	2MLP	2X	38.0	59.7	41.5	22.2	41.6	48.8
Faster	R101v1b-FPN	2MLP	1X	39.9	62.1	43.5	23.1	44.4	51.1
Faster	R101v1b-FPN	2MLP	2X	40.4	62.1	44.0	23.2	44.4	52.7
Faster	R152v1b-FPN	2MLP	1X	41.5	63.5	45.7	24.7	46.0	53.3
Faster	R152v1b-FPN	2MLP	2X	42.0	63.6	45.9	24.8	45.9	55.0
Mask(BBox)	R50v1b-FPN	2MLP	1X	37.8	59.9	40.9	22.9	41.5	48.0
Mask(BBox)	R50v1b-FPN	2MLP	2X	38.6	60.3	41.8	22.6	42.4	49.8
Mask(BBox)	R101v1b-FPN	2MLP	1X	40.4	62.2	44.1	24.0	44.4	52.1
Mask(BBox)	R101v1b-FPN	2MLP	2X	41.3	62.8	45.0	23.9	45.4	53.7
Mask(BBox)	R152v1b-FPN	2MLP	1X	41.8	63.7	46.1	25.3	46.3	53.6
Mask(BBox)	R152v1b-FPN	2MLP	2X	42.8	63.8	46.8	24.6	47.1	55.9
Mask(Inst)	R50v1b-FPN	2MLP	1X	34.4	56.5	36.2	18.7	37.9	46.4
Mask(Inst)	R50v1b-FPN	2MLP	2X	34.9	56.9	37.1	18.3	38.4	47.8
Mask(Inst)	R101v1b-FPN	2MLP	1X	36.3	58.8	38.6	19.4	39.7	49.7
Mask(Inst)	R101v1b-FPN	2MLP	2X	36.9	59.3	39.4	19.1	40.7	51.0
Mask(Inst)	R152v1b-FPN	2MLP	1X	37.4	60.1	39.8	20.0	41.6	50.7
Mask(Inst)	R152v1b-FPN	2MLP	2X	38.0	60.6	40.6	19.8	41.9	52.8
Trident	R50v1b-C4	C5-128ROI	1X	38.4	59.7	41.5	21.4	43.6	52.4
Trident	R50v1b-C4	C5-128ROI	2X	39.6	60.9	42.9	22.5	44.5	53.9
Trident	R101v1b-C4	C5-128ROI	1X	42.2	63.6	45.3	24.5	47.2	57.7
Trident	R101v1b-C4	C5-128ROI	2X	43.0	64.3	46.3	25.3	47.9	58.4
Trident	R152v1b-C4	C5-128ROI	1X	43.7	64.1	48.0	26.9	47.9	58.9
Trident	R152v1b-C4	C5-128ROI	2X	44.4	65.4	48.3	26.4	49.4	59.6
TridentFast	R50v1b-C4	C5-128ROI	1X	37.7	58.7	40.3	19.5	42.4	52.7
TridentFast	R50v1b-C4	C5-128ROI	2X	39.0	60.2	41.8	20.8	43.6	53.8
TridentFast	R101v1b-C4	C5-128ROI	1X	41.1	62.5	43.9	22.1	45.7	57.7
TridentFast	R101v1b-C4	C5-128ROI	2X	42.5	63.7	46.0	23.3	46.7	59.3
TridentFast	R152v1b-C4	C5-128ROI	1X	42.7	64.0	45.6	23.4	47.5	59.1
TridentFast	R152v1b-C4	C5-128ROI	2X	43.9	65.1	47.0	25.1	48.1	60.4
Retina	R50v1b-FPN	4Conv	1X	36.6	56.9	39.0	20.3	40.7	47.2
Retina	R101v1b-FPN	4Conv	1X	39.2	59.5	42.2	22.8	44.0	51.1
Retina	R152v1b-FPN	4Conv	1X	40.4	61.1	43.4	23.6	45.0	52.3
Faster	R50v1b-C4-DCNv1	C5-512ROI	1X	38.8	60.0	41.3	20.6	43.3	53.2
Faster	R101v1b-C4-DCNv1	C5-512ROI	1X	41.4	63.0	44.7	22.7	46.1	56.8
Faster	R50v1b-C4-DCNv2	C5-512ROI	1X	39.6	60.8	42.7	20.8	43.9	54.2
Faster	R50v1b-C4-DCNv2	C5-512ROI	2X	41.2	62.2	44.7	21.7	45.3	57.0
Faster	R101v1b-C4-DCNv2	C5-512ROI	1X	41.7	63.0	44.7	22.8	46.1	57.3
Faster	R101v1b-C4-DCNv2	C5-512ROI	2X	42.7	63.7	46.0	24.9	46.9	57.9
Retina	R50v1b-FPN-TR152v1b1X	4Conv	1X	38.9	59.0	41.6	21.4	43.3	52.1
Retina	R50v1b-FPN-TR152v1b1X	4Conv	2X	40.1	60.6	43.1	21.8	44.5	54.3
Faster	R50v1b-FPN-TR152v1b2X	2MLP	1X	39.9	61.3	43.6	22.7	44.2	52.7
Faster	R50v1b-FPN-TR152v1b2X	2MLP	2X	40.5	62.2	43.9	23.1	44.7	53.9

Box, and Mask Detection Baselines

All AP results are reported on minival2014 of the COCO dataset.

Model	Backbone	Head	Train Schedule	GPU	Image/GPU	FP16	Train MEM	Train Speed	Box AP(Mask AP)	Link
Faster	R50v1-C4	C5-512ROI	1X	8X 1080Ti	2	no	5.9G(4.5G)	20 img/s	34.2	model
Faster	R50v1-C4	C5-512ROI	1X	8X TitanV	2	yes	6.1G	49 img/s	34.4	model
Faster	R50v2-C4	C5-256ROI	1X	8X 1080Ti	2	no	5.1G	33 img/s	32.8	model
Cascade	R50v2-C5	2MLP	1X	8X 1080Ti	2	no	5.9G	25 img/s	38.8	model
Cascade	R50v1-FPN	2MLP	1X	8X 1080Ti	2	no	6.6G	21 img/s	40.3	model
Faster	R50v1-FPN	2MLP	1X	8X 1080Ti	2	no	4.2G(2.6G)	43 img/s	36.5	model
Mask	R50v1-FPN	2MLP+4CONV	1X	8X 1080Ti	2	no	5.7G(3.6G)	35 img/s	37.1(33.7)	model
Retina	R50v1-FPN	4Conv	1X	8X 1080Ti	2	no	4.7G(2.2G)	44 img/s	35.6	model
Trident	R50v2-C4	C5-128ROI	1X	8X 1080Ti	2	no	7.0G(5.3G)	20 img/s	37.1	model
Faster	R101v2-C4	C5-256ROI	1X	8X 1080Ti	2	no	6.7G	25 img/s	37.6	model
Faster-SyncBN	R101v2-C4	C5-256ROI	1X	8X 1080Ti	2	no	7.8G	17 img/s	38.6	model
Faster	R101v1-C4	C5-512ROI	1X	8X 1080Ti	2	no	10.2G	16 img/s	38.3	model
Faster	R101v1-C4	C5-512ROI	1X	8X TitanV	2	yes	7.0G	35 img/s	38.1	model
Faster	R101v1-FPN	2MLP	1X	8X 1080Ti	2	no	5.3G(3.4G)	31 img/s	38.7	model
Cascade	R101v2-C5	2MLP	1X	8X 1080Ti	2	no	7.6G	22 img/s	41.0	model
Cascade	R101v1-FPN	2MLP	1X	8X 1080Ti	2	no	8.7G	19 img/s	42.3	model
Trident	R101v2-C4	C5-128ROI	1X	8X 1080Ti	1	no	6.6G	9 img/s	40.6	model
Trident-Fast	R101v2-C4	C5-128ROI	1X	8X 1080Ti	1	no	6.6G	9 img/s	39.9	model
Retina	R101v1-FPN	4Conv	1X	8X 1080Ti	2	no	5.9G(3.0G)	31 img/s	37.8	model

FP16 Speed Benchmark

Here we provide the FP16 speeed benchmark results of several models.

Model	Backbone	Head	Train Schedule	GPU	Image/GPU	FP16	Train MEM	Train Speed
Faster	R50v1-C4	C5-512ROI	1X	8X 1080Ti	2	no	8.4G	20 img/s
Faster	R50v1-C4	C5-512ROI	1X	8X TitanV	2	yes	6.1G	49 img/s
Faster	R50v1-C4	C5-512ROI	1X	8X TitanV	4	yes	11.2G	55 img/s
Faster	R50v2-C4	C5-256ROI	1X	8X 1080Ti	2	no	5.1G	33 img/s
Faster	R50v2-C4	C5-256ROI	1X	8X TitanV	2	yes	3.8G	61 img/s
Faster	R50v2-C4	C5-256ROI	1X	8X TitanV	4	yes	6.6G	73 img/s
Faster	R101v1-C4	C5-512ROI	1X	8X 1080Ti	2	no	10.2G	16 img/s
Faster	R101v1-C4	C5-512ROI	1X	8X TitanV	2	yes	7.0G	35 img/s
Faster	R50v1-FPN	2MLP	1X	8X 1080Ti	2	no	4.2G(2.6G)	43 img/s
Faster	R50v1-FPN	2MLP	1X	8X 2080Ti	2	yes	3.7G(3.1G)	65 img/s
Faster	R50v1-FPN	2MLP	1X	8X 2080Ti	4	yes	6.2G(6.4G)	77 img/s