findface-extraction-api

The findface-extraction-api service uses neural networks to detect an object in an image, extract the object feature vector, and recognize object attributes (for example, the clothing color for bodies).

It interfaces with the findface-sf-api service as follows:

  • Gets original images with objects and normalized object images.

  • Returns the object bounding box coordinates and, if requested by findface-sf-api, feature vector and object attribute data.

Functionality:

  • object detection in an original image (with a return of the bbox coordinates),

  • object normalization,

  • feature vector extraction from a normalized image,

  • object attribute recognition (a person’s gender, age, emotions; clothing color; car color, car model, etc.).

The findface-extraction-api service can be based on CPU (installed from the docker.int.ntl/ntech/universe/extraction-api-cpu image) or GPU (installed from the docker.int.ntl/ntech/universe/extraction-api-gpu image). For both CPU- and GPU-accelerated services, configuration is done through the findface-extraction-api.yaml configuration file. You can find its default content here for CPU and here for GPU.

When configuring findface-extraction-api (on CPU or GPU), refer to the following parameters:

Parameter

Description

detectors → max_batch_size

Upper limit on detection batch size. When using the CPU, you can specify max_batch_size: -1, then the detector maximum batch size will correspond to the number of CPU cores (default 1).

normalizers → max_batch_size

Upper limit on normalization batch size. When using the CPU, you can specify max_batch_size: -1, then the normalizer maximum batch size will correspond to the number of CPU cores (default 1).

extractors → max_batch_size

Upper limit on extraction batch size. When using the CPU, you can specify max_batch_size: -1, then the extractor maximum batch size will correspond to the number of CPU cores (default 1).

detectors → … → min_object_size

The minimum size of an object (bbox) guaranteed to be detected. The larger the value, the less resources required for object detection.

gpu_device

(Only for GPU) GPU device ID used by findface-extraction-api-gpu.

license_ntls_server

The findface-ntls license server IP address and port.

If necessary, you can also enable recognition models for face attributes, body and body attributes, vehicle and vehicle attributes, and liveness detection. You can find the detailed step-by-step instructions in the following sections:

Important

The acceleration type for each model must match the acceleration type of findface-extraction-api: CPU or GPU. Note that findface-extraction-api on CPU can work only with CPU-models, while findface-extraction-api on GPU supports both CPU- and GPU-models.