Model performance


1
2	* Mean ± 1 standard deviation for model tested on 8 shaves and 2 network inference threads. FPS estimated from throughput of 20 frames over 5 iterations. * near FPS denotes 1 inference thread.
3	** Mean ± 1 standard deviation of 25 FPS computation obtained from the whole pipeline.
4	*** Reported latency or lower (latency includes download of output over USB3 and time for frame to come to the camera - at USB3 speeds 300x300x3 should be below ms), model tested on 8 shaves and 1 network inference thread over 5 iterations
5	° Estimated from OpenVINO workbench.
6	°° Estimated using numpy from .bin file.
7
8	Model Name	Shape	Backbone	Source	NN Only*	NN Only (Latency)***	Whole pipeline with host	GFlops°	MParams°°	ZOO Type	Model name in ZOO
9	Model Name	Shape	Backbone	Source	NN Only*	NN Only (Latency)***	CPU**	GFlops°	MParams°°	ZOO Type	Model name in ZOO
10	EffiecientNet-B0	224 x 224		OpenVINO	28.70 ± 0.1 FPS	60.14 ± 0.02 ms	25.13 ± 1.06 FPS	0.819	5.268	intel	efficientnet-b0
11	ResNet-50	224 x 224		OpenVINO	26.44 ± 0.10 FPS	56.47 ± 0.05 ms	23.40 ± 1.13 FPS	8.2164	25.53	intel	resnet-50-tf
12	MobileNetV2	224 x 224	MobileNetV2	OpenVINO	66.47 ± 0.42 FPS	27.18 ± 0.01 ms	51.86 ± 1.85 FPS	0.876	3.489	intel	mobilenet-v2
13	MobileNetV2 SSD	300 x 300	MobileNetV2	OpenVINO	27.30 ± 0.04 FPS	61.98 ± 0.03 ms	20.33 ± 1.01 FPS	3.775	16.818	intel	ssd_mobilenet_v2_coco
14	Inception V4	299 x 299		TF	8.02 ± 0.03 FPS	167.01 ± 0.07 ms	7.57 ± 0.40 FPS	24.6119	92.018	depthai	inceptionv4_299x299
15	tinyYoloV3	416 x 416	Darknet	OpenVINO	44.83 ± 1.50 FPS	41.05 ± 0.10 ms	41.88 ± 0.23 FPS	5.582	8.848	intel	yolo-v3-tiny-tf
16	OpenPose	456 x 256		OpenVINO	10.54 ± 0.08 FPS	172.65 ± 0.05 ms	8.71 ± 0.39 FPS	60.125	33.165	intel	human-pose-estimation-0001
17	Super Resolution	270 x 480		OpenVINO				11.654	0.03	intel	single-image-super-resolution-1032
18	VGG-19	224 x 224	VGG19	OpenVINO	4.66 ± 0.01 FPS	319.72 ± 0.07 ms	4.79 ± 0.26 FPS	39.3	143.667	intel	vgg19
19	Unet	368 x 480	Unet	OpenVINO	0.45 ± 0.01 FPS*	2223.36 ± 0.03 ms	0.44 ± 0.00 FPS	260.1	31.03	intel	unet-camvid-onnx-0001
20	YOLOP	320 x 320	CSPDarknet	GitHub	20.60 ± 0.09 FPS	87.95 ± 0.03 ms		7.7885	7.936	depthai	yolop_320x320
21	Depth Estimation	240 x 320	MobileNetV2	GitHub	20.03 ± 0.70 FPS	78.07 ± 0.02 ms		8.3246	13.126	depthai	depth_estimation_mbnv2_240x320
22	DM-Count	540 x 960	VGG19	GitHub	0.22 ± 0.01 FPS*	4625.33 ± 0.09 ms		425.1607	21.499	depthai	dmcount_540x960
23	EAST	256 x 256	PVANet	GitHub, PINTO0309	22.50 ± 0.22 FPS	64.21 ± 0.17 ms		9.6895	24.041	depthai	east_text_detection_256x256
24	Fast Depth	256 x 320	MobileNet	GitHub, PINTO0309	40.32 ± 0.14 FPS	46.13 ± 0.06 ms		1.2254	1.352	depthai	fast_depth_256x320
25	Head Pose Estimation	160 x 160	HRNetV2	GitHub, PINTO0309	159.92 ± 0.72 FPS			0.0907	0.626	depthai	head-pose-estimation_160x160
26	HR-Depth	192 x 640	MobileNetV3	GitHub, PINTO0309	17.50 ± 0.02 FPS	107.15 ± 0.11 ms		2.5111	3.099	depthai	hrdepth_192x640
27	Mega Depth	192 x 256	HG	GitHub, PINTO0309	4.80 ± 0.01 FPS	369.50 ± 0.07 ms		60.4724	5.357	depthai	megadepth
28	Mobile Object Localizer	192 x 192	MobileNetV2	TF	63.72 ± 0.33 FPS	24.51 ± 0.01 ms			1.744	depthai	mobile_object_localizer_192x192
29	SC-Depth	256 x 832	DispResNet	GitHub	12.89 ± 0.33 FPS	135.79 ± 0.03 ms		17.1095	14.322	depthai	scdepth_256x832
30	MicroNet-m0	224 x 224		GitHub	34.36 ± 0.04 FPS	33.35 ± 0.01 ms	30.51 ± 0.98 FPS	0.00884	1.018	depthai	micronet_m0_224x224
31	GhostNet	320 x 256		GitHub	53.13 ± 0.92 FPS	34.62 ± 0.01 ms	44.04 ± 1.66 FPS	0.47133	5.179	depthai	ghostnet_320x256
32	ShuffleNetv2	224 x 224		GitHub	151.72 ± 2.55 FPS	10.26 ± 0.02 ms	85.71 ± 2.21 FPS	0.29572	2.271	depthai	shufflenetv2_224x224
33	YoloV3	416 x 416	DarkNet	GitHub	4.16 ± 0.14 FPS	399.21 ± 0.06 ms		65.99438	61.923	depthai	yolov3_coco_416x416
34	YoloV4-tiny	416 x 416	DarkNet	GitHub	39.80 ± 1.33 FPS	43.70 ± 0.03 ms		6.9294	6.054	depthai	yolov4_tiny_coco_416x416
35	YoloV4	608 x 608	DarkNet	GitHub	1.31 ± 0.01 FPS	1185.57 ± 0.22 ms		129.55669	64.33	depthai	yolov4_coco_608x608
36	Mask R-CNN	300 x 300	ResNet-50	GitHub	3.11 ± 0.01 FPS	528.82 ± 0.03 ms		57.35542	46.804	depthai	mask_rcnn_resnet50_coco_300x300
37	7 Segment recognition	256 x 750			18.48 ± 0.05 FPS	91.82 ± 0.01 ms				depthai	7_segment_recognition_256x750
38	Cone Slayer	416 x 416	YoloV7tiny	GitHub	53.97 ± 0.03 FPS	33.30 ± 0.02 ms				depthai	coneslayer_416x416
39	Depth Estimation	480 x 640	MobileNetV2	GitHub	4.75 ± 0.08 FPS	356.01 ± 0.03 ms				depthai	depth_estimation_mbnv2_480x640
40	Face recognition ArcFace	112 x 112	MobileFaceNet	GitHub	126.63 ± 0.09 FPS	13.65 ± 0.00 ms				depthai	face-recognition-arcface-112x112
41	Face detector YUNET	160 x 120	YuNet	GitHub	363.61 ± 0.83 FPS	4.68 ± 0.01 ms				depthai	face_detection_yunet_160x120
42	Facemesh	192 x 192		GitHub	143.39 ± 0.28 FPS	11.85 ± 0.01 ms				depthai	facemesh_192x192
43	Facial Landmarks	160 x 160	HRNetV2	GitHub	153.85 ± 0.45 FPS	8.04 ± 0.01 ms				depthai	facial_landmarks_68_160x160
44	Fast Depth	480 x 640	MobileNet	GitHub	10.79 ± 0.03 FPS	171.63 ± 0.04 ms				depthai	fast_depth_480x640
45	Image Quality	256 x 256	EdgeSegNet		26.81 ± 0.07 FPS	70.89 ± 0.01 ms				depthai	image_quality_assessment_256x256
46	MobileNetV2	224 x 224		PyTorch Hub	86.39 ± 0.30 FPS	20.70 ± 0.02 ms				depthai	mobilenetv2_imagenet_embedder_224x224
47	Padim Wood	256 x 256		GitHub	11.40 ± 0.18 FPS	147.08 ± 0.06 ms				depthai	padim_wood_256x256
48	Palm detector	128 x 128		GitHub	69.82 ± 0.52 FPS	22.30 ± 0.02 ms				depthai	palm_detection_128x128
49	Person reid	96 x 48	RMNet	OpenVINO	294.57 ± 3.34 FPS	5.13 ± 0.00 ms				depthai	person-reidentification-retail-0031_96x48
50	QR Code detector	384 x 384		GitHub	100.78 ± 1.29 FPS	15.69 ± 0.01 ms				depthai	qr_code_detection_384x384
51	SBD Mask	224 x 224		GitHub	103.04 ± 0.30 FPS	17.59 ± 0.00 ms				depthai	sbd_mask_classification_224x224
52	DeepLab V3	256 x 256		GitHub	36.53 ± 0.64 FPS	48.12 ± 0.03 ms				depthai	deeplab_v3_mnv2_256x256
53	DeepLab V3	513 x 513		GitHub	6.32 ± 0.09 FPS	253.12 ± 0.04 ms				depthai	deeplab_v3_mnv2_513x513
54	YoloV5n	416 x 416	Darknet	GitHub	41.10 ± 0.21 FPS	43.23 ± 0.07 ms				depthai	yolov5n_coco_416x416
55	YoloV6n R1	416 x 416	EfficientRep	GitHub	67.13 ± 0.09 FPS	28.57 ± 0.02 ms				depthai	yolov6n_coco_416x416
56	YoloV6n R1	640 x 640	EfficientRep	GitHub	29.96 ± 0.14 FPS	65.27 ± 0.04 ms				depthai	yolov6n_coco_640x640
57	YoloV6t R1	416 x 416	EfficientRep	GitHub	23.76 ± 0.67 FPS	77.18 ± 0.02 ms				depthai	yolov6t_coco_416x416
58	YoloV7	416 x 416	E-ELAN	GitHub	3.61 ± 0.01 FPS	409.98 ± 0.06 ms				depthai	yolov7_coco_416x416
59	YoloV7t	416 x 416	E-ELAN	GitHub	47.42 ± 0.55 FPS	37.50 ± 0.06 ms				depthai	yolov7tiny_coco_416x416
60	MobileOne S0	224x224		GitHub	165.45 ± 0.22 FPS	11.09 ± 0.01 ms				None
61	Resnet18	224x225		PyTorch Hub	94.78 ± 0.25 FPS	19.71 ± 0.01 ms				None
62	YoloV6n R2	416x416	EfficientRep	GitHub	65.51 ± 0.06 FPS	29.28 ± 0.02 ms				None
63	YoloV6n R2	640x640	EfficientRep	GitHub	29.30 ± 0.16 FPS	66.42 ± 0.03 ms				None
64	YoloV6t R2	416x416	EfficientRep	GitHub	35.78 ± 0.06 FPS	54.09 ± 0.05 ms				None
65	YoloV6t R2	640x640	EfficientRep	GitHub	14.16 ± 0.04 FPS	133.60 ± 0.06ms				None
66	YoloV6s R2	416x416	EfficientRep	GitHub	20.63 ± 0.28 FPS	89.95 ± 0.02 ms				None
67	YoloV6s R2	640x640	EfficientRep	GitHub	6.69 ± 0.01 FPS	247.18 ± 0.07 ms				None
68	YoloV6m R2	416x416	EfficientRep	GitHub	8.57 ± 0.03 FPS	190.24 ± 0.07 ms				None
69	YoloV6m R2	640x640	EfficientRep	GitHub	3.40 ± 0.00 FPS	487.37 ± 0.10 ms				None
70	YoloV6l R2	416x416	EfficientRep	GitHub	3.58 ± 0.15 FPS	450.06 ± 0.05 ms				None
71	YoloV6l R2	640x640	EfficientRep	GitHub	1.45 ± 0.00 FPS	1090.63 ± 0.12 ms				None
72	YoloV6n R2.1	416x416	EfficientRep	GitHub	65.82 ± 0.27 FPS	29.27 ± 0.02 ms				None
73	YoloV6n R2.1	640x640	EfficientRep	GitHub	29.25 ± 0.08 FPS	66.40 ± 0.01 ms				None
74	YoloV6s R2.1	416x416	EfficientRep	GitHub	27.51 ± 0.07 FPS	68.62 ± 0.01 ms				None
75	YoloV6s R2.1	640x640	EfficientRep	GitHub	9.54 ± 0.21 FPS	177.50 ± 0.02 ms				None
76	YoloV6m R2.1	416x416	EfficientRep	GitHub	10.25 ± 0.06 FPS	164.47 ± 0.04 ms				None
77	YoloV6m R2.1	640x640	EfficientRep	GitHub	4.19 ± 0.03 FPS	418.89 ± 0.11 ms				None
78	YoloV6l R2.1	416x416	EfficientRep	GitHub	4.67 ± 0.30 FPS	340.72 ± 0.06 ms				None
79	YoloV6l R2.1	640x640	EfficientRep	GitHub	1.78 ± 0.07 FPS	867.46 ± 0.11 ms				None
80	YoloV6n R3	416x416	EfficientRep	GitHub	50.45 ± 0.04 FPS	37.76 ± 0.03 ms				None
81	YoloV6n R3	640x640	EfficientRep	GitHub	23.43 ± 0.04 FPS	82.05 ± 0.01 ms				None
82	YoloV6s R3	416x416	EfficientRep	GitHub	19.50 ± 0.02 FPS	96.00 ± 0.01 ms				None
83	YoloV6s R3	640x640	EfficientRep	GitHub	6.51 ± 0.01 FPS	259.71 ± 0.12 ms				None
84	YoloV6m R3	416x416	EfficientRep	GitHub	7.82 ± 0.04 FPS	201.36 ± 0.04 ms				None
85	YoloV6m R3	640x640	EfficientRep	GitHub	3.27 ± 0.00 FPS	508.29 ± 0.03 ms				None
86	YoloV6l R3	416x416	EfficientRep	GitHub	3.31 ± 0.09 FPS	475.65 ± 0.06 ms				None
87	YoloV6l R3	640x640	EfficientRep	GitHub	1.38 ± 0.00 FPS	1141.58 ± 0.05 ms				None
88	YoloV7t	416x416	E-ELAN	GitHub	46.68 ± 0.48 FPS	37.55 ± 0.06 ms				None
89	YoloV7t	640x640	E-ELAN	GitHub	17.80 ± 0.61 FPS	97.03 ± 0.18 ms				None
90	YoloV7	416x416	E-ELAN	GitHub	3.61 ± 0.01 FPS	410.03 ± 0.16 ms				None
91	YoloV7	640x640	E-ELAN	GitHub	1.43 ± 0.00 FPS	1009.25 ± 0.14 ms				None
92	YoloV7x	416x416	E-ELAN	GitHub	2.20 ± 0.00 FPS	749.29 ± 0.10 ms				None
93	YoloV7x	640x640	E-ELAN	GitHub	0.57 ± 0.00 FPS	1749.53 ± 0.13 ms				None
94	YoloV8n	416x416		GitHub	31.30 ± 0.16 FPS	56.94 ± 0.02 ms				None
95	YoloV8n	640x640		GitHub	14.28 ± 0.04 FPS	123.60 ± 0.03 ms				None
96	YoloV8s	416x416		GitHub	15.17 ± 0.17 FPS	111.87 ± 0.03 ms				None
97	YoloV8s	640x640		GitHub	5.81 ± 0.07 FPS	280.10 ± 0.04 ms				None
98	YoloV8m	416x416		GitHub	5.97 ± 0.02 FPS	273.77 ± 0.05 ms				None
99	YoloV8m	640x640		GitHub	2.71 ± 0.01 FPS	636.13 ± 0.03 ms				None
100	YoloV8l	416x416		GitHub	2.67 ± 0.02 FPS	601.94 ± 0.08 ms				None
101	YoloV8l	640x640		GitHub	1.28 ± 0.00 FPS	1169.15 ± 0.06 ms				None
102	YoloV8x	416x416		GitHub	2.11 ± 0.00 FPS	790.16 ± 0.06 ms				None
103	YoloV8x	640x640		GitHub	0.68 ± 0.00 FPS	2097.45 ± 0.17 ms				None
104
105	LEGEND
106	Classification
107	Object detection
108	Depth Estimation
109	Segmentation
110	Embedder