person-vehicle-bike-detection-crossroad-1016

Use Case and High-Level Description

MobileNetV2 + SSD-based network is for Person/Vehicle/Bike detection in security surveillance applications. Works in a variety of scenes and weather/lighting conditions.

Example

Specification

Metric Value
Mean Average Precision (mAP) 62.55%
AP people 73.63%
AP vehicles 77.84%
AP bikes 36.18%
Max objects to detect 200
GFlops 3.560
Source framework PyTorch*

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
Non-vehicle 62,643

Similarly, training dataset has 219,181 images with:

Type of object Number of bounding boxes
Vehicle 810,323
Pedestrian 1,114,799
Non-vehicle 62,334

Inputs

Image, name: input.1, shape: 1, 3, 512, 512 in the format B, C, H, W, where:

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

Expected color order: BGR.

Outputs

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 (0 - non-vehicle, 1 - vehicle, 2 - person)
  • 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.