11. 模型性能Benchmark

11.1. 说明

  • 测试条件：

    • 测试开发板：J5-DVB board。

    • 测试核心数：latency单核，fps双核。

    • 性能数据获取频率设置为：5分钟时间内性能参数的平均值。

    • Python版本：Python3.8。

  • 缩写说明：

    • C = 计算量，单位为GOPs（十亿次运算/秒）。此数据通过 hb_perf 工具获得。

    • FPS = 每秒帧率。此数据在开发板单线程运行ai_benchmark_j5示例包/script路径下各模型子文件夹的 fps.sh 脚本获取，包含后处理。

    • ITC = 推理耗时，单位为ms（毫秒）。此数据在开发板单线程运行ai_benchmark_j5示例包/script路径下各模型子文件夹的 latency.sh 脚本获取，不含后处理。

    • TCPP = 后处理耗时，单位为ms（毫秒）。此数据在开发板单线程运行ai_benchmark_j5示例包/script路径下各模型子文件夹的 latency.sh 脚本获取。

    • RV = 单帧读取数据量，单位为mb（兆比特）。此数据通过 hb_perf 工具获得。

    • WV = 单帧写入数据量，单位为mb（兆比特）。此数据通过 hb_perf 工具获得。

11.2. 模型重要性能数据

MODEL NAME

INPUT SIZE

C(GOPs)

FPS

ITC(ms)

TCPP(ms)

ACCURACY

Dataset

MobileNetv1

1x224x224x3

1.14

3795.89

0.912

0.062

Top1:

0.7061(FLOAT)

0.7026(INT8)

ImageNet

MobileNetv2

1x224x224x3

0.63

4195.81

0.777

0.090

Top1:

0.7265(FLOAT)

0.7153(INT8)

ImageNet

GoogleNet

1x224x224x3

3.00

2322.02

1.209

0.062

Top1:

0.7001(FLOAT)

0.6989(INT8)

ImageNet

Resnet18

1x224x224x3

3.65

1559.31

1.656

0.061

Top1:

0.6837(FLOAT)

0.6829(INT8)

ImageNet

EfficientNet_Lite0

1x224x224x3

0.77

2794.87

1.111

0.062

Top1:

0.7490(FLOAT)

0.7469(INT8)

ImageNet

EfficientNet_Lite1

1x240x240x3

1.20

2399.65

1.763

0.062

Top1:

0.7648(FLOAT)

0.7624(INT8)

ImageNet

EfficientNet_Lite2

1x260x260x3

1.72

2136.43

1.329

0.062

Top1:

0.7738(FLOAT)

0.7715(INT8)

ImageNet

EfficientNet_Lite3

1x280x280x3

2.77

1604.74

1.637

0.061

Top1:

0.7922(FLOAT)

0.7902(INT8)

ImageNet

EfficientNet_Lite4

1x300x300x3

5.11

1070.20

2.259

0.062

Top1:

0.8069(FLOAT)

0.8058(INT8)

ImageNet

Vargconvnet

1x224x224x3

9.06

1573.61

1.620

0.062

Top1:

0.7790(FLOAT)

0.7785(INT8)

ImageNet

Efficientnasnet_m

1x300x300x3

4.53

1142.02

2.108

0.062

Top1:

0.7973(FLOAT)

0.7916(INT8)

ImageNet

Efficientnasnet_s

1x280x280x3

1.44

2705.96

1.097

0.062

Top1:

0.7578(FLOAT)

0.7518(INT8)

ImageNet

YOLOv2_Darknet19

1x608x608x3

62.94

280.03

7.262

1.678

[IoU=0.50:0.95]=

0.2760(FLOAT)

0.2730(INT8)

COCO

YOLOv3_Darknet53

1x416x416x3

65.90

209.00

9.797

9.954

[IoU=0.50:0.95]=

0.3330(FLOAT)

0.3350(INT8)

COCO

YOLOv5x_v2.0

1x672x672x3

243.91

77.08

25.417

30.846

[IoU=0.50:0.95]=

0.4800(FLOAT)

0.4660(INT8)

COCO

Ssd_mobilenetv1

1x300x300x3

2.30

2589.66

1.104

1.102

mAP:

0.7342(FLOAT)

0.7275(INT8)

VOC

Centernet_resnet101

1x512x512x3

90.54

250.32

8.307

4.665

[IoU=0.50:0.95]=

0.3420(FLOAT)

0.3350(INT8)

COCO

YOLOv3_VargDarknet

1x416x416x3

42.82

301.08

6.847

9.956

[IoU=0.50:0.95]=

0.3350(FLOAT)

0.3270(INT8)

COCO

Deeplabv3plus_efficientnetb0

1x1024x2048x3

30.78

203.61

10.005

0.775

mIoU:

0.7630(FLOAT)

0.7568(INT8)

Cityscapes

Fastscnn_efficientnetb0

1x1024x2048x3

12.49

294.37

7.140

0.778

mIoU:

0.6997(FLOAT)

0.6928(INT8)

Cityscapes

Deeplabv3plus_efficientnetm1

1x1024x2048x3

77.05

117.59

17.001

0.770

mIoU:

0.7794(FLOAT)

0.7740(INT8)

Cityscapes

Deeplabv3plus_efficientnetm2

1x1024x2048x3

124.16

89.94

22.403

0.775

mIoU:

0.7882(FLOAT)

0.7856(INT8)

Cityscapes

Resnet50

1x224x224x3

7.72

683.26

3.149

0.089

Top1:

0.7737(FLOAT)

0.7674(INT8)

ImageNet

VargNetV2

1x224x224x3

0.72

3524.78

0.859

0.090

Top1:

0.7394(FLOAT)

0.7321(INT8)

ImageNet

Swint

1x224x224x3

8.98

138.65

14.705

0.089

Top1:

0.8024(FLOAT)

0.7947(INT8)

ImageNet

MixVarGENet

1x224x224x3

2.07

5823.47

0.649

0.089

Top1:

0.7133(FLOAT)

0.7066(INT8)

ImageNet

Fcos_efficientnetb0

1x512x512x3

5.02

1757.32

1.465

0.247

[IoU=0.50:0.95]=

0.3626(FLOAT)

0.3562(INT8)

COCO

Fcos_efficientnetb2

1x768x768x3

22.08

451.58

4.834

6.700

[IoU=0.50:0.95]=

0.4470(FLOAT)

0.4470(INT8)

COCO

Fcos_efficientnetb3

1x896x896x3

41.45

267.65

7.797

9.047

[IoU=0.50:0.95]=

0.4720(FLOAT)

0.4740(INT8)

COCO

Pointpillars_kitti_car

1x1x150000x4

66.82

116.47

32.161

2.571

APDet=

0.7731(FLOAT)

0.7676(INT8)

Kitti3d

RetinaNet_vargnetv2_fpn

1x1024x1024x3

301.27

80.93

24.647

6.535

[IoU=0.50:0.95]=

0.3151(FLOAT)

0.3129(INT8)

COCO

Yolov3_mobilenetv1

1x416x416x3

20.58

491.90

4.346

1.658

mAP:

0.7657(FLOAT)

0.7581(INT8)

VOC

Ganet_mixvargenet

1x320x800x3

10.74

2443.55

1.141

0.956

F1Score:

0.7949(FLOAT)

0.7872(INT8)

CuLane

DETR_resnet50

1x800x1333x3

202.99

47.41

41.370

1.675

[IoU=0.50:0.95]=

0.3570(FLOAT)

0.3134(INT8)

MS COCO

DETR_efficientnetb3

1x800x1333x3

67.31

62.28

32.346

1.670

[IoU=0.50:0.95]=

0.3721(FLOAT)

0.3597(INT8)

MS COCO

FCOS3D_efficientnetb0

1x512x896x3

19.94

605.57

3.979

8.747

NDS:

0.3062(FLOAT)

0.3019(INT8)

nuscenes

Centerpoint_pointpillar

300000x5

127.73

101.15

24.597

52.534

NDS:

0.5832(FLOAT)

0.5814(INT8)

nuscenes

Keypoint_efficientnetb0

1x128x128x3

0.45

3251.61

0.908

0.361

PCK(alpha=0.1):

0.9433(FLOAT)

0.9431(INT8)

carfusion

Unet_mobilenetv1

1x1024x2048x3

7.36

1047.20

2.107

0.589

mIoU:

0.6802(FLOAT)

0.6753(INT8)

Cityscapes

Pwcnet_pwcnetneck

1x384x512x6

81.71

161.24

12.682

0.307

EndPointError:

1.4117(FLOAT)

1.4075(INT8)

flyingchairs

Motr_efficientnetb3

image:

1x800x1422x3

track_query:

1x2x128x156

ref_points:

1x2x128x4

mask_query:

1x1x256x1

64.43

67.06

26.533

23.247

MOTA:

0.5802(FLOAT)

0.5776(INT8)

Mot17

Bev_lss_efficientnetb0_multitask

image:

6x256x704x3

points(0&1):

10x128x128x2

2.41

278.52

7.936

17.898

NDS:

0.3006(FLOAT)

0.3000(INT8)

MeanIOU:

0.5180(FLOAT)

0.5148(INT8)

nuscenes

Bev_gkt_mixvargenet_multitask

image:

6x512x960x3

points(0-8):

6x64x64x2

34.49

85.77

23.687

17.926

NDS:

0.2809(FLOAT)

0.2791(INT8)

MeanIOU:

0.4851(FLOAT)

0.4836(INT8)

nuscenes

Bev_ipm_efficientnetb0_multitask

image:

6x512x960x3

points:

6x128x128x2

8.83

208.78

9.715

17.990

NDS:

0.3053(FLOAT)

0.3041(INT8)

MeanIOU:

0.5146(FLOAT)

0.5099(INT8)

nuscenes

Bev_ipm_4d_efficientnetb0_multitask

image:

6x512x960x3

points:

6x128x128x2

prev_feat:

1x128x128x64

prev_point:

1x128x128x2

8.93

187.69

10.559

18.112

NDS:

0.3724(FLOAT)

0.3725(INT8)

MeanIOU:

0.5290(FLOAT)

0.5388(INT8)

nuscenes

Detr3d_efficientnetb3_nuscenes

coords(0-3):

6x4x256x2

image:

6x512x1408x3

masks:

1x4x256x24

37.55

27.08

69.307

2.415

NDS:

0.3304(FLOAT)

0.3283(INT8)

nuscenes

Petr_efficientnetb3_nuscenes

image:

6x512x1408x3

pos_embed:

1x96x44x256

36.24

8.41

226.051

2.420

NDS:

0.3760(FLOAT)

0.3733(INT8)

nuscenes

Centerpoint_mixvargnet_multitask

300000x5

51.45

103.99

23.544

50.460

NDS:

0.5809(FLOAT)

0.5762(INT8)

MeanIOU:

0.9129(FLOAT)

0.9122(INT8)

nuscenes

Stereonetplus_mixvargenet

2x544x960x3

24.29

244.63

6.386

15.406

EPE:

1.1270(FLOAT)

1.1352(INT8)

SceneFlow

Densetnt_vectornet

goals_2d:

30x1x2048x2

goals_2d_mask:

30x1x2048x1

instance_mask:

30x1x96x1

lane_feat:

30x9x64x11

traj_feat:

30x19x32x9

0.42

86.93

26.634

10.623

minFDA:

1.2974(FLOAT)

1.3038(INT8)

Argoverse 1

11.3. 模型全部性能数据

11.3.1. MobileNetv1

  • INPUT SIZE: 1x224x224x3

  • C(GOPs): 1.14

  • FPS: 3795.89

  • ITC(ms): 0.912

  • TCPP(ms): 0.062

  • RV(mb): 3.89

  • WV(mb): 0.02

  • Dataset: ImageNet

  • ACCURACY: Top1: 0.7061(FLOAT)/0.7026(INT8)

  • LINKS: https://github.com/shicai/MobileNet-Caffe

11.3.2. MobileNetv2

  • INPUT SIZE: 1x224x224x3

  • C(GOPs): 0.63

  • FPS: 4195.81

  • ITC(ms): 0.777

  • TCPP(ms): 0.090

  • RV(mb): 2.94

  • WV(mb): 0.02

  • Dataset: ImageNet

  • ACCURACY: Top1: 0.7265(FLOAT)/0.7153(INT8)

11.3.3. GoogleNet

11.3.4. Resnet18

11.3.5. EfficientNet_Lite0

11.3.6. EfficientNet_Lite1

11.3.7. EfficientNet_Lite2

11.3.8. EfficientNet_Lite3

11.3.9. EfficientNet_Lite4

11.3.10. Vargconvnet

11.3.11. Efficientnasnet_m

11.3.12. Efficientnasnet_s

11.3.13. YOLOv2_Darknet19

  • INPUT SIZE: 1x608x608x3

  • C(GOPs): 62.94

  • FPS: 280.03

  • ITC(ms): 7.262

  • TCPP(ms): 1.678

  • RV(mb): 47.35

  • WV(mb): 2.43

  • Dataset: COCO

  • ACCURACY: [IoU=0.50:0.95]= 0.2760(FLOAT)/0.2730(INT8)

  • LINKS: https://pjreddie.com/darknet/yolo

11.3.14. YOLOv3_Darknet53

  • INPUT SIZE: 1x416x416x3

  • C(GOPs): 65.90

  • FPS: 209.00

  • ITC(ms): 9.797

  • TCPP(ms): 9.954

  • RV(mb): 57.75

  • WV(mb): 4.94

  • Dataset: COCO

  • ACCURACY: [IoU=0.50:0.95]= 0.3330(FLOAT)/0.3350(INT8)

  • LINKS: https://github.com/ChenYingpeng/caffe-yolov3

11.3.15. YOLOv5x_v2.0

11.3.16. Ssd_mobilenetv1

  • INPUT SIZE: 1x300x300x3

  • C(GOPs): 2.30

  • FPS: 2589.66

  • ITC(ms): 1.104

  • TCPP(ms): 1.102

  • RV(mb): 5.82

  • WV(mb): 0.20

  • Dataset: VOC

  • ACCURACY: mAP: 0.7342(FLOAT)/0.7275(INT8)

  • LINKS: https://github.com/chuanqi305/MobileNet-SSD

11.3.17. Centernet_resnet101

11.3.18. YOLOv3_VargDarknet

11.3.19. Deeplabv3plus_efficientnetb0

11.3.20. Fastscnn_efficientnetb0

11.3.21. Deeplabv3plus_efficientnetm1

11.3.22. Deeplabv3plus_efficientnetm2

11.3.23. Resnet50

  • INPUT SIZE: 1x224x224x3

  • C(GOPs): 7.72

  • FPS: 683.26

  • ITC(ms): 3.149

  • TCPP(ms): 0.089

  • RV(mb): 24.03

  • WV(mb): 0.52

  • Dataset: ImageNet

  • ACCURACY: Top1: 0.7737(FLOAT)/0.7674(INT8)

11.3.24. VargNetV2

  • INPUT SIZE: 1x224x224x3

  • C(GOPs): 0.72

  • FPS: 3524.78

  • ITC(ms): 0.859

  • TCPP(ms): 0.090

  • RV(mb): 3.68

  • WV(mb): 0.03

  • Dataset: ImageNet

  • ACCURACY: Top1: 0.7394(FLOAT)/0.7321(INT8)

11.3.25. Swint

  • INPUT SIZE: 1x224x224x3

  • C(GOPs): 8.98

  • FPS: 138.65

  • ITC(ms): 14.705

  • TCPP(ms): 0.089

  • RV(mb): 40.95

  • WV(mb): 1.31

  • Dataset: ImageNet

  • ACCURACY: Top1: 0.8024(FLOAT)/0.7947(INT8)

11.3.26. MixVarGENet

  • INPUT SIZE: 1x224x224x3

  • C(GOPs): 2.07

  • FPS: 5823.47

  • ITC(ms): 0.649

  • TCPP(ms): 0.089

  • RV(mb): 2.26

  • WV(mb): 0.02

  • Dataset: ImageNet

  • ACCURACY: Top1: 0.7133(FLOAT)/0.7066(INT8)

11.3.27. Fcos_efficientnetb0

  • INPUT SIZE: 1x512x512x3

  • C(GOPs): 5.02

  • FPS: 1757.32

  • ITC(ms): 1.465

  • TCPP(ms): 0.247

  • RV(mb): 4.73

  • WV(mb): 0.30

  • Dataset: COCO

  • ACCURACY: [IoU=0.50:0.95]= 0.3626(FLOAT)/0.3562(INT8)

11.3.28. Fcos_efficientnetb2

11.3.29. Fcos_efficientnetb3

11.3.30. Pointpillars_kitti_car

  • INPUT SIZE: 1x1x150000x4

  • C(GOPs): 66.82

  • FPS: 116.47

  • ITC(ms): 32.161

  • TCPP(ms): 2.571

  • RV(mb): 43.30

  • WV(mb): 24.47

  • Dataset: Kitti3d

  • ACCURACY: APDet= 0.7731(FLOAT)/0.7676(INT8)

11.3.31. RetinaNet_vargnetv2_fpn

  • INPUT SIZE: 1x1024x1024x3

  • C(GOPs): 301.27

  • FPS: 80.93

  • ITC(ms): 24.647

  • TCPP(ms): 6.535

  • RV(mb): 74.94

  • WV(mb): 37.52

  • Dataset: COCO

  • ACCURACY: [IoU=0.50:0.95]= 0.3151(FLOAT)/0.3129(INT8)

11.3.32. Yolov3_mobilenetv1

  • INPUT SIZE: 1x416x416x3

  • C(GOPs): 20.58

  • FPS: 491.90

  • ITC(ms): 4.346

  • TCPP(ms): 1.658

  • RV(mb): 25.87

  • WV(mb): 1.53

  • Dataset: VOC

  • ACCURACY: mAP: 0.7657(FLOAT)/0.7581(INT8)

11.3.33. Ganet_mixvargenet

  • INPUT SIZE: 1x320x800x3

  • C(GOPs): 10.74

  • FPS: 2443.55

  • ITC(ms): 1.141

  • TCPP(ms): 0.956

  • RV(mb): 1.36

  • WV(mb): 0.21

  • Dataset: CuLane

  • ACCURACY: F1Score: 0.7949(FLOAT)/0.7872(INT8)

11.3.34. DETR_resnet50

  • INPUT SIZE: 1x800x1333x3

  • C(GOPs): 202.99

  • FPS: 47.41

  • ITC(ms): 41.370

  • TCPP(ms): 1.675

  • RV(mb): 174.56

  • WV(mb): 100.56

  • Dataset: MS COCO

  • ACCURACY: [IoU=0.50:0.95]= 0.3570(FLOAT)/0.3134(INT8)

11.3.35. DETR_efficientnetb3

  • INPUT SIZE: 1x800x1333x3

  • C(GOPs): 67.31

  • FPS: 62.28

  • ITC(ms): 32.346

  • TCPP(ms): 1.670

  • RV(mb): 119.12

  • WV(mb): 63.56

  • Dataset: MS COCO

  • ACCURACY: [IoU=0.50:0.95]= 0.3721(FLOAT)/0.3597(INT8)

11.3.36. FCOS3D_efficientnetb0

  • INPUT SIZE: 1x512x896x3

  • C(GOPs): 19.94

  • FPS: 605.57

  • ITC(ms): 3.979

  • TCPP(ms): 8.747

  • RV(mb): 11.55

  • WV(mb): 5.10

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.3062(FLOAT)/0.3019(INT8)

11.3.37. Centerpoint_pointpillar

  • INPUT SIZE: 300000x5

  • C(GOPs): 127.73

  • FPS: 101.15

  • ITC(ms): 24.597

  • TCPP(ms): 52.534

  • RV(mb): 39.37

  • WV(mb): 19.04

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.5832(FLOAT)/0.5814(INT8)

11.3.38. Keypoint_efficientnetb0

  • INPUT SIZE: 1x128x128x3

  • C(GOPs): 0.45

  • FPS: 3251.61

  • ITC(ms): 0.908

  • TCPP(ms): 0.361

  • RV(mb): 4.41

  • WV(mb): 0.04

  • Dataset: carfusion

  • ACCURACY: PCK(alpha=0.1): 0.9433(FLOAT)/0.9431(INT8)

11.3.39. Unet_mobilenetv1

  • INPUT SIZE: 1x1024x2048x3

  • C(GOPs): 7.36

  • FPS: 1047.20

  • ITC(ms): 2.107

  • TCPP(ms): 0.589

  • RV(mb): 6.96

  • WV(mb): 2.88

  • Dataset: Cityscapes

  • ACCURACY: mIoU: 0.6802(FLOAT)/0.6753(INT8)

11.3.40. Pwcnet_pwcnetneck

  • INPUT SIZE: 1x384x512x6

  • C(GOPs): 81.71

  • FPS: 161.24

  • ITC(ms): 12.682

  • TCPP(ms): 0.307

  • RV(mb): 27.65

  • WV(mb): 15.32

  • Dataset: flyingchairs

  • ACCURACY: EndPointError: 1.4117(FLOAT)/1.4075(INT8)

11.3.41. Motr_efficientnetb3

  • INPUT SIZE: image: 1x800x1422x3 track_query: 1x2x128x156 ref_points: 1x2x128x4 mask_query: 1x1x256x1

  • C(GOPs): 64.43

  • FPS: 67.06

  • ITC(ms): 26.533

  • TCPP(ms): 23.247

  • RV(mb): 73.95

  • WV(mb): 28.09

  • Dataset: Mot17

  • ACCURACY: MOTA: 0.5802(FLOAT)/0.5776(INT8)

11.3.42. Bev_lss_efficientnetb0_multitask

  • INPUT SIZE: image: 6x256x704x3 points(0&1): 10x128x128x2

  • C(GOPs): 2.41

  • FPS: 278.52

  • ITC(ms): 7.936

  • TCPP(ms): 17.898

  • RV(mb): 2.56

  • WV(mb): 1.99

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.3006(FLOAT)/0.3000(INT8) MeanIOU: 0.5180(FLOAT)/0.5148(INT8)

11.3.43. Bev_gkt_mixvargenet_multitask

  • INPUT SIZE: image: 6x512x960x3 points(0-8): 6x64x64x2

  • C(GOPs): 34.49

  • FPS: 85.77

  • ITC(ms): 23.687

  • TCPP(ms): 17.926

  • RV(mb): 10.98

  • WV(mb): 6.89

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.2809(FLOAT)/0.2791(INT8) MeanIOU: 0.4851(FLOAT)/0.4836(INT8)

11.3.44. Bev_ipm_efficientnetb0_multitask

  • INPUT SIZE: image: 6x512x960x3 points: 6x128x128x2

  • C(GOPs): 8.83

  • FPS: 208.78

  • ITC(ms): 9.715

  • TCPP(ms): 17.990

  • RV(mb): 5.14

  • WV(mb): 3.63

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.3053(FLOAT)/0.3041(INT8) MeanIOU: 0.5146(FLOAT)/0.5099(INT8)

11.3.45. Bev_ipm_4d_efficientnetb0_multitask

  • INPUT SIZE: image: 6x512x960x3 points: 6x128x128x2 prev_feat: 1x128x128x64 prev_point: 1x128x128x2

  • C(GOPs): 8.93

  • FPS: 187.69

  • ITC(ms): 10.559

  • TCPP(ms): 18.112

  • RV(mb): 5.70

  • WV(mb): 3.99

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.3724(FLOAT)/0.3725(INT8) MeanIOU: 0.5290(FLOAT)/0.5388(INT8)

11.3.46. Detr3d_efficientnetb3_nuscenes

  • INPUT SIZE: coords(0-3): 6x4x256x2 image: 6x512x1408x3 masks: 1x4x256x24

  • C(GOPs): 37.55

  • FPS: 27.08

  • ITC(ms): 69.307

  • TCPP(ms): 2.415

  • RV(mb): 57.77

  • WV(mb): 40.23

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.3304(FLOAT)/0.3283(INT8)

11.3.47. Petr_efficientnetb3_nuscenes

  • INPUT SIZE: image: 6x512x1408x3 pos_embed: 1x96x44x256

  • C(GOPs): 36.24

  • FPS: 8.41

  • ITC(ms): 226.051

  • TCPP(ms): 2.420

  • RV(mb): 301.52

  • WV(mb): 167.79

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.3760(FLOAT)/0.3733(INT8)

11.3.48. Centerpoint_mixvargnet_multitask

  • INPUT SIZE: 300000x5

  • C(GOPs): 51.45

  • FPS: 103.99

  • ITC(ms): 23.544

  • TCPP(ms): 50.460

  • RV(mb): 33.84

  • WV(mb): 14.45

  • Dataset: nuscenes

  • ACCURACY: NDS: 0.5809(FLOAT)/0.5762(INT8) MeanIOU: 0.9129(FLOAT)/0.9122(INT8)

11.3.49. Stereonetplus_mixvargenet

  • INPUT SIZE: 2x544x960x3

  • C(GOPs): 24.29

  • FPS: 244.63

  • ITC(ms): 6.386

  • TCPP(ms): 15.406

  • RV(mb): 7.02

  • WV(mb): 6.74

  • Dataset: SceneFlow

  • ACCURACY: EPE: 1.1270(FLOAT)/1.1352(INT8)

11.3.50. Densetnt_vectornet

  • INPUT SIZE: goals_2d: 30x1x2048x2 goals_2d_mask: 30x1x2048x1 instance_mask: 30x1x96x1 lane_feat: 30x9x64x11 traj_feat: 30x19x32x9

  • C(GOPs): 0.42

  • FPS: 86.93

  • ITC(ms): 26.634

  • TCPP(ms): 10.623

  • RV(mb): 3.29

  • WV(mb): 2.92

  • Dataset: Argoverse 1

  • ACCURACY: minFDA: 1.2974(FLOAT)/1.3038(INT8)