Use Case and High-Level Description

Person/Vehicle/Bike detector is based on SSD detection architecture, RMNet backbone, and learnable image downscale block (like person-vehicle-bike-detection-crossroad-0066, but with extra pooling). The model is intended for security surveillance applications and works in a variety of scenes and weather/lighting conditions.



Metric Value
Mean Average Precision (mAP) 65.12%
AP people 77.47%
AP vehicles 74.94%
AP bikes 44.14%
Max objects to detect 200
GFlops 3.964
MParams 1.178
Source framework Caffe*

Average Precision (AP) is defined as an area under the precision/recall curve.

Validation dataset consists of 34,757 images from various scenes and includes:

Type of object Number of bounding boxes
Vehicle 229,503
Pedestrian 240,009
Bike 62,643

Similarly, training dataset has 160,297 images with:

Type of object Number of bounding boxes
Vehicle 501,548
Pedestrian 706,786
Bike 55,692


Image, name: input, shape: 1, 3, 1024, 1024 in the format B, C, H, W, where:

  • B - batch size
  • C - number of channels
  • H - image height
  • W - image width

The expected color order is BGR.


The net outputs blob with shape: 1, 1, 200, 7 in the format 1, 1, N, 7, where N is the number of detected bounding boxes. Each detection has the format [image_id, label, conf, x_min, y_min, x_max, y_max], where:

  • image_id - ID of the image in the batch
  • label - predicted class ID (1- person, 2 - vehicle, 3 - bike)
  • conf - confidence for the predicted class
  • (x_min, y_min) - coordinates of the top left bounding box corner
  • (x_max, y_max) - coordinates of the bottom right bounding box corner

Legal Information

[*] Other names and brands may be claimed as the property of others.