Abstract: A sensor apparatus includes a cylindrical sensor window defining an axis oriented substantially vertically, an air nozzle positioned below the sensor window and aimed upward, and a sensor-housing-top cap positioned above the sensor window and including a sensor-housing top and a circumferential cap. The sensor-housing-top cap is a single piece. The circumferential cap includes a topside and an underside. The topside and the underside are disposed radially outward from the sensor window. The underside includes a groove having a cross-section elongated circumferentially around the axis. The cross-section of the groove curves from a lower end upwardly and outwardly to an upper end.

Abstract: A facial images retrieval system is provided. The facial images retrieval system is adapted to receive an initial textual description of a facial image to perform an initial facial image search that obtains a plurality of facial images based on the textual description. The facial images retrieval system then receives a selection of the first and second facial images that are relatively close to a desired facial image to perform a further facial image search to obtain another facial image.

Abstract: A sensor apparatus includes a cylindrical sensor window defining an axis oriented substantially vertically, an air nozzle positioned below the sensor window and aimed upward, and a sensor-housing-top cap positioned above the sensor window and including a sensor-housing top and a circumferential cap. The sensor-housing-top cap is a single piece. The circumferential cap includes a topside and an underside. The topside and the underside are disposed radially outward from the sensor window. The underside includes a groove having a cross-section elongated circumferentially around the axis. The cross-section of the groove curves from a lower end upwardly and outwardly to an upper end.

Abstract: A computer vision neural network system is provided. The computer vision neural network system includes a computer with a multi-attribute loss module that concatenates first and second normalized feature vector descriptors to generate a master feature vector descriptor that describes attributes and classes of a facial image. The multi-attribute loss module estimates an error distance between the master feature vector descriptor and a plurality of class center vector descriptors. Each class center vector descriptor is a mean of a plurality of master feature vector descriptors associated with a soft-biometric class in a plurality of faces in a plurality of facial images.

Abstract: A computer vision neural network system is provided. The computer vision neural network system includes a computer with a multi-attribute loss module that concatenates first and second normalized feature vector descriptors to generate a master feature vector descriptor that describes attributes and classes of a facial image. The multi-attribute loss module estimates an error distance between the master feature vector descriptor and a plurality of class center vector descriptors. Each class center vector descriptor is a mean of a plurality of master feature vector descriptors associated with a soft-biometric class in a plurality of faces in a plurality of facial images.

Abstract: A facial images retrieval system is provided. The facial images retrieval system is adapted to receive an initial textual description of a facial image to perform an initial facial image search that obtains a plurality of facial images based on the textual description. The facial images retrieval system then receives a selection of the first and second facial images that are relatively close to a desired facial image to perform a further facial image search to obtain another facial image.