The sample marked as 🚧 is not provided by MNN and is not guaranteed to be available. If it is not available, please leave a message in the MNN DingTalk group.

DeepLab

Sample: Android

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
deeplabv3_257_mv_gpu 38.02 46.08

DenseNet

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
densenet 110.58 179.51

Inception

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
inception_v3 285.02 278.58
inception_v3_quant 302.50 210.19
inception_v4 801.81 1414.61
inception_resnet_v2 523.33 723.45

LaneNet

示例：Linux/MacOS/Ubuntu(Lane Detection)🚧by MaybeShewill

🏷TensorFlow

Model / Code tag 0.2.1.5 iPhone 6s Plus MI 6 (4 threads)
lanenet_model 1878.01

LFFD

Sample：Linux/MacOS/Ubuntu(face detection)🚧by 哈喽

🏷MXNet

Model / Code tag 0.2.0.9 iPhone 6s Plus MI 6 (4 threads)
symbol_10_560_25L_8scales_v1 302.08 (input size = 640x480)
symbol_10_320_20L_5scales_v2 198.68 (input size = 640x480)

MnasNet

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
mnasnet_1.3_224 34.70 62.73

MobileNet

Sample: iOS, Android, Android(文档检测)🚧by

🏷Caffe

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
mobilenet 27.29 29.96
mobilenet_v2 27.11 30.23

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
mobilenet_v1_1.0_224 27.82 29.00
mobilenet_v1_1.0_224_quant 30.16 14.01
mobilenet_v2_1.0_224 22.01 22.86
mobilenet_v2_1.0_224_quant 24.06 17.16

MobileNet SSD

Sample: Android(vehicle/pedestrain detection)🚧by 张新栋

🏷Caffe

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
mobilenet_ssd 47.84 52.63

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
ssd_mobilenet_v1_coco 49.49 53.24
ssd_mobilenet_v2_coco 66.73 72.18

Modified MobileNet SSD (Ultra Light Fast Generic Face Detector ≈1MB)

Sample：Linux🚧by

🏷Caffe

Model / Code tag 0.2.1.5 iPhone 6s Plus 小米6 (4 threads)
RFB-320-quant-ADMM-32 10.25
RFB-320-quant-KL-5792 10.44
RFB-320 7.80
slim-320-quant-ADMM-50 9.04
slim-32 6.33

MTCNN

Sample：AIoT(license plate recognize)🚧by 仪山湖, Android(face detection)🚧by

🏷Caffe

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
mtcnn_v1_det1 0.02 0.04
mtcnn_v1_det2 0.19 0.23
mtcnn_v1_det3 0.80 0.85
mtcnn_v2_det1 0.03 0.04
mtcnn_v2_det2 0.19 0.25
mtcnn_v2_det3 0.71 0.83
mtcnn_v2_det4 1.14 1.34

Multi Person MobileNet

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
multi_person_mobilenet_v1_075 33.39 41.01

SqueezeNet

Sample: iOS, Android

🏷Caffe

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
squeezenet_v1.0 (batch = 10) 374.63 418.06
squeezenet_v1.1 (batch = 10) 186.66 187.51

🏷TensorFlow

Model / Code tag 0.2.0.8 iPhone 6s Plus MI 6 (4 threads)
squeezenet_tf 33.99 65.47

YOLO(s)

Sample：Linux/MacOS/Ubuntu(object detection)🚧by Linux(YOLO sets) 🚧by david8862

🏷TensorFlow

Model / Code tag 0.2.0.9 iPhone 6s Plus MI 6 (4 threads)
voc320_opti_fc_quant 608.46
voc544_opti_fc 1979.78