Appendix B. Glossary#
Glossary
| Term | Description |
|---|---|
| Host memory | Computer system RAM |
| Device memory | On-board RAM of GPU or NPU card |
| Memory transfer | Operation that copies memory from host to device or vice-versa |
Descriptor#
A set of features meant to describe a real-world object (e.g., a person's face). Computed by means of computer vision algorithms, such features are typically matched to each other to determine the similarity of represented objects.
Cooperative Photoshooting and Recognition#
A procedure of taking person face photograph characterized by person awareness of the matter and his/her will to assist.
Typical highlights:
- Close to frontal head pose;
- Neutral facial expression;
- No occlusions (i.e., hair, hats, non-transparent eyewear, hands, other objects obscuring the face);
- No extreme lighting conditions (i.e., reasonable illuminance, no direct sunlight);
- Steady and well-tuned optics (i.e., no motion blur, depth of field, digital post-processing except noise cancellation).
Cooperative photoshooting is opposite to the so-called "in the wild" photoshooting, which is also called non-cooperative shooting (or recognition).
Matching#
The process of descriptors comparison. Matching is usually implemented as a distance function applied to the feature sets and distances comparison later on. The smaller the distance, the closer are descriptors, hence, the more similar are the objects.
For convenience, helper functions exist to convert distance to a normalized similarity score, where 100% means completely identical, and 0% means completely different.