Face Detection Clause Samples
Face Detection. This module has a lightweight face detector tailored for DNN inference and can be applied to any live experience that requires an accurate facial region of interest as an input for other task-specific models, such as facial key point estimation, facial features or expression classification, and face region segmentation. (Price: USD 2M)
Face Detection. FDR (TE-204) – UPDATE The face detection extractor was deployed in June 2015 and subsequent releases. It uses the open source software libccv which implements the SURF-Cascade Detection [LZ13]. The evaluation report will describe how the training model will perform on the showcase evaluation. In order to decrease the amount of annotation data as well as the processing time in the video analysis case, we chose two pipeline-based approaches:
1. Face detection is run on frames at fixed time steps.
2. Face detection is run on frames extracted as shot boundaries or key frames from the Temporal Video Segmentation Extractor. The pipeline section (Section 2.3) gives more information about this. However, the first approach will be more usable once the new broker version supports job-specific parametrization of the extractor from a user perspective. For Year 3, there are no major updates planned for this extractor. If required an additional face model could be trained if there is a usable annotated data set available for this purpose.