# mobilenet-yolo-v4-syg¶

## Use Case and High-Level Description¶

This is a Keras* version of mobilenet-yolov4 model designed to perform real-time vehicle detection. The weights are pretrained by BDD100k and retrained by our own dataset. For details, see the repository, paper of MobileNetV2 and YOLOv4.

Metric

Value

Type

Detection

GFLOPs

65.984

MParams

61.922

Source framework

Keras*

## Accuracy¶

Accuracy metrics obtained on SYGDate0829 ‘SYGDate0829.z01’’SYGDate0829.z02’’SYGDate0829.z03’’SYGDate0829.zip’ which is our own made* validation dataset for converted model.

Metric

Value

mAP

86.35%

## Input¶

### Original model¶

Image, name - input_1, shape - 1, 416, 416, 3, format is B, H, W, C where:

• B - batch size

• H - height

• W - width

• C - channel

Channel order is RGB. Scale value - 255.

### Converted model¶

Image, name - input_1, shape - 1, 416, 416, 3, format is B, H, W, C where:

• B - batch size

• H - height

• W - width

• C - channel

Channel order is BGR.

## Output¶

### Original model¶

1. The array of detection summary info, name - separable_conv2d_22/BiasAdd, shape - 1,52,52,27. The anchor values are 12,16, 19,36, 40,28.

2. The array of detection summary info, name - separable_conv2d_30/BiasAdd, shape - 1,26,26,27. The anchor values are 36,75, 76,55, 72,146.

3. The array of detection summary info, name - separable_conv2d_38/BiasAdd, shape - 1,13,13,27. The anchor values are 142,110, 192,243, 459,401.

For each case format is B,Cx,Cy,N\*14,, where

• B - batch size

• Cx, Cy - cell index

• N - number of detection boxes for cell

Detection box has format [x, y, h, w, box_score, class_no_1, …, class_no_4], where:

• (x, y) - raw coordinates of box center, apply sigmoid function to get relative to the cell coordinates

• h, w - raw height and width of box, apply exponential function and multiply by corresponding anchors to get absolute height and width values

• box_score - confidence of detection box, apply sigmoid function to get confidence in [0,1] range

• class_no_1,…, class_no_4 - probability distribution over the classes in logits format, apply sigmoid function and multiply by obtained confidence value to get confidence of each class

### Converted model¶

1. The array of detection summary info, name - separable_conv2d_22/separable_conv2d/YoloRegion, shape - 1, 52, 52, 27. The anchor values are 12,16, 19,36, 40,28.

2. The array of detection summary info, name - separable_conv2d_30/separable_conv2d/YoloRegion, shape - 1, 26, 26, 27. The anchor values are 36,75, 76,55, 72,146.

3. The array of detection summary info, name - separable_conv2d_38/separable_conv2d/YoloRegion, shape - 1, 13, 13, 27. The anchor values are 142,110, 192,243, 459,401.

Detection box has format [x, y, h, w, box_score, class_no_1, …, class_no_4], where:

• (x, y) - coordinates of box center relative to the cell

• h, w - raw height and width of box, apply exponential function and multiply by corresponding anchors to get absolute height and width values

• box_score - confidence of detection box in [0,1] range

• class_no_1,…, class_no_4 - probability distribution over the classes in the [0,1] range, multiply by confidence value to get confidence of each class

You can download models and if necessary convert them into OpenVINO™ IR format using the Model Downloader and other automation tools as shown in the examples below.

omz_downloader --name <model_name>
omz_converter --name <model_name>