OpenCV 5 DNN Benchmarks

This page holds the complete CPU inference benchmarks for the new OpenCV 5.x DNN engine, compared against ONNX Runtime. For the summary and context, see the Performance section of the OpenCV 5 page.

How to read these tables

• All numbers are inference time in milliseconds (lower is better).
• Columns:
  • OpenCV — the new OpenCV 5.x DNN engine.
  • ONNX Runtime (1.25.1) — the official ONNX Runtime binaries published on the project's GitHub releases page.
• error marks a model that failed to run in that configuration; — marks a measurement that is not yet available.

---

Intel i9

Model	OpenCV	ONNX Runtime (1.25.1)
BiRefNet	7178	9503.1
BlazeFace	1.11	1.03
depth_anything	135.35	126.5
Dinov2	21	26.5
Dinov3	16	14.5
NAFNet	1826	1518.85
NanoTrack	0.67	0.71
owlv2	1074.73	1542.6
ppocr	152.9	161.8
RAFT	1193	685.2
RealEsrGAN	488	580
RetinaFACE	23.63	21
RTMPose	6.1	4.8
SAM2 ENCODER	2350.57	2280.78
SAM2 DECODER	22.77	10.94
SAM3 (Encoder)	7753.16	6194.47
SAMURAI (Encoder)	1104.8	1074.7
SAMURAI (Decoder)	22	30.7
Segformer	172	152.6
SigLip	23.67	26.28
SwinIR	225.7	164.7
trOCR (Encoder)	120	106.8
trOCR (Decoder)	47.3	38.9
U2 Net	105	100
X-Feat	6.2	10.3
yolo26m_seg	110.2	105
yolo26n_seg	17	15.9
yolo26n	9.9	9.6
yunet	4.04	2.34
RF-DETR	93.3	97.4
RT-DETR	118.6	99.9
Grounding DINO	1845	1862
Fast Neural Style	14.86	40.89
AlexNet_ONNX	6.67	6.23
GoogLeNet_ONNX	4.04	4.58
SqueezeNet_ONNX	1.12	0.86
DenseNet121_ONNX	15.14	10.92
Inception_ONNX	3.37	4.27
ResNet_18_v1	3.46	3.06
ResNet_50_v1	7.28	6.78
ResNet50_QDQ	4.39	5.53
MobileNetv2	1.35	1.06
EfficientNet	6.32	3.49
VIT_B_32	13.99	12.57
VIT_Base_Patch16_224	38.2	35.68
DeiT_Tiny_Patch16_224	6.41	4.89
MobileViT_XS	5.81	5.39
MobileViTv2_100	7.07	7.09
BEiT_Base_Patch16_224	42.36	35.83
YOLOv3	13.35	14.78
YOLOv4	106.09	179.31
YOLOv4_tiny	6.94	6.47
YOLOv5	9.39	8.67
YOLOv8	12.42	12.98
YOLOX	24.16	23.8
MobileNet_SSD_v1_ONNX	10.56	7.15
SFace	1.92	3.87
MPPalm	1.9	2.12
MPHand	1.94	1.32
MPPose	6.17	2.79
PPHumanSeg	2.16	2.45
CRNN	15.56	1.79
VitTrack	1.31	1.04
BERT	11.33	11.71
FacePaint	422.39	478.35

---

Intel i7

Model	OpenCV	ONNX Runtime (1.25.1)
BiRefNet	14649.64	18964.991
BlazeFace	1.657	0.9
depth_anything	324.342	332.495
Dinov2	40.463	72.477
Dinov3	34.924	27.967
NAFNet	5231.817	3424.311
NanoTrack	1.71	1.12
owlv2	2761.63	3809.29
ppocr	340.8	235
RAFT	error	1041.5
RealEsrGAN	1664.339	1394.293
RetinaFACE	73.02	77.54
RTMPose	15.723	12.87
SAM2 ENCODER	5669.38	7261.39
SAM2 DECODER	52.03	30.29
SAM3 (Encoder)	20133.3	15859.25
SAMURAI (Encoder)	2457.7	2222.2
SAMURAI (Decoder)	48.3	24.1
Segformer	503.91	453.72
SigLip	91.35	72.51
SwinIR	688.242	485.465
trOCR (Encoder)	234.3	217.3
trOCR (Decoder)	94.4	75.9
U2 Net	306.478	249.562
X-Feat	14.358	22.187
yolo26m_seg	373.5	391.9
yolo26n_seg	43.7	43.4
yolo26n	24.9	19
yunet	10.7	4.68
RF-DETR	226.6	227.4
RT-DETR	313.1	243.8
Grounding DINO	4277.66	4459.09
Fast Neural Style	43.759	75.517
AlexNet_ONNX	14.08	14.88
GoogLeNet_ONNX	10.44	11.59
SqueezeNet_ONNX	3.58	1.6
DenseNet121_ONNX	35.03	17.71
Inception_ONNX	11.02	10.91
ResNet_18_v1	10.34	6.15
ResNet_50_v1	18.51	13.28
ResNet50_QDQ	64.39	11.79
MobileNetv2	3.51	2.11
EfficientNet	14.03	7.01
VIT_B_32	33.83	33.12
VIT_Base_Patch16_224	76	69.1
DeiT_Tiny_Patch16_224	11.02	7.4
MobileViT_XS	12.13	10.32
MobileViTv2_100	17.09	13.55
BEiT_Base_Patch16_224	80.89	85.87
YOLOv3	38.8	31.09
YOLOv4	331.39	794.21
YOLOv4_tiny	20.18	21.35
YOLOv5	22.86	16.44
YOLOv8	28.82	25.54
YOLOX	68.33	83.79
MobileNet_SSD_v1_ONNX	29.75	15.64
SFace	5.1	9.15
MPPalm	4.46	3.4
MPHand	4.39	1.27
MPPose	11.73	5.37
PPHumanSeg	6.43	3.95
CRNN	33.08	2.99
VitTrack	3.38	1.1
BERT	23.92	18.98
FacePaint	877.13	1221.76

---

Apple M1

Model	OpenCV	ONNX Runtime (1.25.1)
BiRefNet	32483.861	26350.284
BlazeFace	1.755	0.669
depth_anything	398.525	554.114
Dinov2	98.259	112.244
Dinov3	86.7	82.6
NAFNet	3730	3809.2
NanoTrack	2.4	2.32
owlv2	7635.68	9513.23
ppocr	323.7	273.9
RAFT	643.9	—
RealEsrGAN	1868.913	2771.789
RetinaFACE	166.67	217.91
RTMPose	33.3	45.1
SAM2 ENCODER	7454.963	8555.085
SAM2 DECODER	42.826	25.444
SAM3 (Encoder)	error	7222.96
SAMURAI (Encoder)	2717.2	3376.2
SAMURAI (Decoder)	43.5	28.7
Segformer	448.4	523
SigLip	125.5	130.6
SwinIR	994	1006.5
trOCR (Encoder)	353.4	409
trOCR (Decoder)	95.7	107.3
U2 Net	816.039	1005.142
X-Feat	38.21	36.017
yolo26m_seg	974.3	950.1
yolo26n_seg	85.6	107.5
yolo26n	55.6	63.4
yunet	11.95	12.82
RF-DETR	297.8	301.8
RT-DETR	408.4	363.8
Grounding DINO	10319.89	3941.3
Fast Neural Style	49.761	71.228
AlexNet_ONNX	10.29	11.89
GoogLeNet_ONNX	15.05	17.52
SqueezeNet_ONNX	3.8	3.53
DenseNet121_ONNX	38.23	32.36
Inception_ONNX	14.79	16.95
ResNet_18_v1	13.25	18.18
ResNet_50_v1	27.58	37.19
ResNet50_QDQ	16.52	11.19
MobileNetv2	5.61	6.16
EfficientNet	20.67	23.19
VIT_B_32	38.11	44.76
VIT_Base_Patch16_224	126.94	143.61
DeiT_Tiny_Patch16_224	16.62	13.33
MobileViT_XS	16.23	16.6
MobileViTv2_100	22.34	23.33
BEiT_Base_Patch16_224	131.79	147.36
YOLOv3	51.76	72.7
YOLOv4	482.87	1001.12
YOLOv4_tiny	26.76	37.3
YOLOv5	34.35	34.26
YOLOv8	41.13	54.71
YOLOX	106.91	144.73
MobileNet_SSD_v1_ONNX	21.34	17.86
SFace	5.68	7.71
MPPalm	5.87	7.08
MPHand	5.35	4.11
MPPose	10.71	6.17
PPHumanSeg	7.85	8.46
CRNN	29.3	7.71
VitTrack	2.71	1.45
BERT	16.84	17.77
FacePaint	1079.25	1231.11

---

Apple M5

Model	OpenCV	ONNX Runtime (1.25.1)
BiRefNet	13746.997	15144.133
BlazeFace	1.406	1.39
depth_anything	220.546	228.585
Dinov2	37.745	38.512
Dinov3	25.51	25.33
NAFNet	2581.816	2649.351
NanoTrack	1.54	1.61
owlv2	2181.32	2427.6
ppocr	139.3	139.5
RAFT	4336	4415.1
RealEsrGAN	1138.484	1210.61
RetinaFACE	45.8	45.86
RTMPose	12.572	12.479
SAM2 ENCODER	3900.4	4072.64
SAM2 DECODER	27.18	27.11
Segformer	292.8	297.31
SigLip	81.73	81.61
SwinIR	351.618	398.654
trOCR (Encoder)	204	209.4
trOCR (Decoder)	58.9	60.1
U2 Net	309.222	299.946
X-Feat	24.095	24.149
yolo26m_seg	252.1	252.1
yolo26n_seg	31.7	31.7
yolo26n	22.7	21.6
yunet	5.18	5.17
RF-DETR	213.3	205.4
RT-DETR	302.8	301.9
Grounding DINO	7697.07	7740.43
Fast Neural Style	31.426	32.235
AlexNet_ONNX	5.85	5.96
GoogLeNet_ONNX	11.4	11
SqueezeNet_ONNX	3.47	3.83
DenseNet121_ONNX	39.85	39.75
Inception_ONNX	10.41	10.04
ResNet_18_v1	7.71	7.75
ResNet_50_v1	16.05	16.07
ResNet50_QDQ	11.27	11.25
MobileNetv2	4.96	4.91
EfficientNet	19.89	20.06
VIT_B_32	28.13	28.2
VIT_Base_Patch16_224	76.68	76.86
DeiT_Tiny_Patch16_224	16.94	17.16
MobileViT_XS	13.57	14.07
MobileViTv2_100	16.59	16.58
BEiT_Base_Patch16_224	82.97	83.49
YOLOv3	27.26	27.31
YOLOv4	238.16	241.8
YOLOv4_tiny	14.87	14.73
YOLOv5	22.48	22.4
YOLOv8	24.25	24.91
YOLOX	54.38	54.33
MobileNet_SSD_v1_ONNX	19.05	18.8
SFace	3.88	3.95
MPPalm	5.02	5
MPHand	4.94	5.03
MPPose	9.65	9.69
PPHumanSeg	6.15	5.98
CRNN	24.06	24.14
VitTrack	2.36	2.33
BERT	13.78	13.54
FacePaint	604.23	613.01

---

AMD

Model	OpenCV	ONNX Runtime (1.25.1)
BiRefNet	5812.648	9830.294
BlazeFace	0.94	3.21
depth_anything	93.037	240.39
Dinov2	13.761	45.186
Dinov3	14.313	14.594
NAFNet	2009.587	2794.632
NanoTrack	0.88	0.78
owlv2	716.71	1722.6
ppocr	158.7	207
RAFT	—	—
RealEsrGAN	561.708	1162.763
RetinaFACE	23.23	27.24
RTMPose	5.828	7.347
SAM2 ENCODER	1813.17	2331.81
SAM2 DECODER	18.44	12.99
SAM3 (Encoder)	7619.52	—
SAMURAI (Encoder)	979.3	1246.1
SAMURAI (Decoder)	19.8	16.3
Segformer	170.09	307.94
SigLip	31.11	73.06
SwinIR	249.553	270.99
trOCR (Encoder)	—	—
trOCR (Decoder)	—	—
U2 Net	102.212	161.191
X-Feat	6.226	11.463
yolo26m_seg	110.8	154.9
yolo26n_seg	20.5	40.6
yolo26n	13.6	30.7
yunet	4.87	5.16
RF-DETR	73.5	176.9
RT-DETR	117.9	154.1
Grounding DINO	1690.48	2986.07
Fast Neural Style	15.449	27.634
AlexNet_ONNX	6.21	13.45
GoogLeNet_ONNX	3.82	8.66
SqueezeNet_ONNX	1.42	2.62
DenseNet121_ONNX	22.41	27.09
Inception_ONNX	3.59	6.76
ResNet_18_v1	3.25	5.11
ResNet_50_v1	8.1	6.73
ResNet50_QDQ	3.69	7.13
MobileNetv2	2.13	3.48
EfficientNet	10.02	7.26
VIT_B_32	11.93	17.29
VIT_Base_Patch16_224	27.69	55.4
DeiT_Tiny_Patch16_224	5.21	10.4
MobileViT_XS	6.45	14.86
MobileViTv2_100	7.92	15.72
BEiT_Base_Patch16_224	29.48	62.29
YOLOv3	13.72	24.78
YOLOv4	121.24	338.87
YOLOv4_tiny	7.48	12.94
YOLOv5	12.03	28.4
YOLOv8	14.65	21.95
YOLOX	25.37	38.93
MobileNet_SSD_v1_ONNX	14.27	14.87
SFace	1.82	3.43
MPPalm	2.04	8.7
MPHand	2.65	2.43
MPPose	6.01	10.81
PPHumanSeg	2.45	5.02
CRNN	12.28	2.62
VitTrack	1.36	3.18
BERT	10.95	25.74
FacePaint	347.4	652.3

---

Back to the OpenCV 5 release notes.