Model List and Performance Benchmarks

Classification

network	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
resnet50_imagenet	77.37	76.99	76.94	ImageNet	1x3x224x224	0.587	6540.931	0.904	1468.542	2.652	299.014
efficientnet_imagenet	74.31	74.23	74.18	ImageNet	1x3x224x224	0.332	19776.52	0.41	4867.534	0.924	316.801
mixvargenet_imagenet	71.33	71.23	71.04	ImageNet	1x3x224x224	0.313	20277.93	0.366	5372.038	0.875	325.121
henet_tinye_imagenet	77.68	77.22	76.92	ImageNet	1x3x224x224	0.381	11482.78	0.498	3618.033	1.074	1631.96
henet_tinym_imagenet	78.38	77.95	77.62	ImageNet	1x3x224x224	0.393	11256.83	0.518	3380.954	1.136	290.305

Detection

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
fcos_efficientnetb3_mscoco	efficientnetb3	48.03	47.65	47.58	MS COCO	1x3x896x896	2.922	1462.401	4.049	265.949	13.884	21.721
deform_detr_resnet50_mscoco	resnet50	44.34	44.65	44.80	MS COCO	1x3x800x1333	77.347	29.45	6.336	158.829	-	-

Segmentation

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
unet_mobilenetv1_cityscapes	MobileNetV1	68.02	67.56	67.53	Cityscapes	1x3x1024x2048	0.877	5726.862	1.198	1146.047	3.869	331.277

3D检测

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
fcos3d_efficientnetb0_nuscenes	efficientnetb0	30.60	30.27	30.31	nuscenes	1x3x512x896	1.461	3375.2	2.139	562.283	6.491	38.403

Bev

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
bev_lss_efficientnetb0_multitask_nuscenes	efficientnetb0	30.06(NDS) 20.62(MAP) 51.80(MeanIOU)	30.10(NDS) 20.51(MAP) 51.78(MeanIOU)	30.08(NDS) 20.46(MAP) 51.47(MeanIOU)	nuscenes det && seg	6x3x256x704, 10x128x128x2, 10x128x128x2	3.254	1341.371	4.597	234.133	20.6883	92.45
detr3d_efficientnetb3_nuscenes	efficientnetb3	33.04(NDS) 27.52(MAP)	32.84(NDS) 27.14(MAP)	32.81(NDS) 27.06(MAP)	nuscenes det	6x3x512x1408	15.4735	236.345	23.069	44.061	75.8594	13.232
petr_efficientnetb3_nuscenes	efficientnetb3	37.65(NDS) 30.38(MAP)	37.26(NDS) 29.29(MAP)	37.40(NDS) 29.33(MAP)	nuscenes det	6x3x512x1408	21.835	180.757	34.279	29.532	104.5687	9.593
bevformer_tiny_resnet50_detection_nuscenes	resnet50	37.12(NDS) 26.79(MAP)	37.16(NDS) 26.50(MAP)	37.15(NDS) 26.59(MAP)	nuscenes det	6x3x480x800, 1x2500x256, 1x50x50x2, 6x20x32x2, 1x100x50x2, 6x640x4x2, 1x2500x1	14.577	270.031	23.043	44.148
bev_sparse_henet_tinym_nuscenes	henet_tinym	54.19(NDS) 43.38(MAP)	52.23(NDS) 42.17(MAP)	-	nuscenes det	6x3x256x704, 6x4x4, 1x384x11, 1x384x256	8.217	483.07	11.814	86.818	29.914	34.778
bev_sparse_det_maptr_flashocc_henet_tinym_nuscenes	henet_tinym_imagenet	52.34(det NDS) 41.33(det mAP) 59.58(map mAP) 31.89 (occ, MIOU)	52.14(det NDS) 41.00(det mAP) 59.06(map mAP) 33.19 (occ, MIOU)	-	nuscenes det && map && occ3d	6x3x256x704, 6x4x4, 1x384x11, 1x384x256, 10x128x128x2, 10x128x128x2	20.47	163.21	31.185	32.519	11.5765	193.1247

Online Map Construction

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6E bpu latency (ms)	J6E FPS	J6B bpu latency (ms)	J6B FPS
maptroe_henet_tinym_bevformer_nuscenes	henet_tinym_imagenet	0.6632	0.6577	0.6567	nuscenes	6x3x480x800, 1x1x50x100, 6x20x100x2, 1x100x100x2, 6x2000x4x2, 1x5000x1	7.499	521.61	10.696	96.247	34.28	29.894
maptroe_sparse_henet_tinym_nuscenes	henet_tinym_imagenet	0.5982	0.5999	0.5992	nuscenes	6x3x256x704, 6x4x4	9.19	258.936	11.553	88.767	29.227	35.199

Occupancy Prediction

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6E bpu latency (ms)	J6E FPS	J6B bpu latency (ms)	J6B FPS
flashocc_henet_lss_occ3d_nuscenes	henet_tinym_imagenet	0.3674	0.3657	0.3693	occ3d_nuscenes	6x3x512x960, 10x128x128x2, 10x128x128x2	8.5437	469.977	10.581	97.12	29.333	35.005

Multiple Object Track

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
motr_efficientnetb3_mot17	efficientnetb3	58.02	57.62	57.76	Mot17	1x3x800x1422, 1x256x2x128, 1x1x1x256, 1x4x2x128	6.545	604.402	9.247	111.782	27.872	36.806

Trajectory Prediction

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
qcnet_oe_argoverse2	-	80.09	79.54	78.21	argoverse 2	输入见下方list	2.842	1572.123	3.888	285.868	12.659	155.818

Note

The indicator for QCNet is the HitRate item.

The input shape of the qcnet_oe_argoverse2 model is:

1x30x10, 1x10x30x30, 1x30x1, 1x1x30x1, 1x1x30x1, 1x1x30x1, 1x1x30x1, 1x1x30x80, 1x1x30x80, 1x1x30x80, 1x1x30x6, 1x1x30x6, 1x1x30x6, 1x1x30x6, 1x1x30x30, 1x1x30x30, 1x1x30x30, 1x30x2x128, 1x30x6x128, 1x80, 1x1x80x80, 1x1x80x80, 1x1x80x80, 1x1x80x50, 1x1x80x50, 1x1x80x50, 1x80x80, 1x80x50, 1x30x30, 1x30x1, 1x80x80

Lidar

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6M bpu latency (ms)	J6M FPS	J6B bpu latency (ms)	J6B FPS
pointpillars_kitti_car	SequentialBottleNeck	77.31	76.86	76.76	KITTI3D	150000x4	21.115	327.568	21.496	197.207	1652.419	2.013
centerpoint_pointpillar_nuscenes	SequentialBottleNeck	58.32(NDS) 48.04(MAP)	58.11(NDS) 47.85(MAP)	58.14(NDS) 47.81(MAP)	nuscenes det	1x5x20x40000, 40000x4	7.957	796.232	9.483	167.473	38.271	60.003

Lidar Fusion

network	backbone	float	qat	quantization	dataset	input shape	J6P bpu latency (ms)	J6P FPS	J6E bpu latency (ms)	J6E FPS	J6B bpu latency (ms)	J6B FPS
bevfusion_pointpillar_henet_multisensor_multitask_nuscenes	henet_tinym_imagenet	64.28(NDS) 58.09(MAP) 51.77(MIOU)	62.91(NDS) 57.48(MAP) 52.51(MIOU)	-	nuscenes det && occ3d	1x5x20x40000, 40000x4, 6x3x512x960, 1x256x128x2, 6x5120x2x2, 1x16384x1	19.841	233.49	26.618	43.573	91.564	15.304
bev_sparse_lidar_fusion_henet_tinym_nuscenes	henet_tinym_imagenet	66.64(NDS) 61.16(MAP)	66.31(NDS), 60.70(mAP)	65.96(NDS) 60.35(mAP)	nuscenes det	6x3x256x704, 6x4x4, 1x384x11, 1x384x256, 1x5x20x40000, 40000x4	17.314	293.429	23.373	51.739	85.034	19.393

#Model List and Performance Benchmarks

#Classification

#Detection

#Segmentation

#3D检测

#Bev

#Online Map Construction

#Occupancy Prediction

#Multiple Object Track

#Trajectory Prediction

#Lidar

#Lidar Fusion