could you give some details about redaction demo dataset ? Is it publicly open or custom created one ? What advices could you give us during data preparation ? How inference of both face and license plate works perfectly in a single caffe model ?
You will need to create a multi-class detector for both face and license detection.
We have some multi-class detector in Deepstream SDK, which can be retrained with TLToolkit directly: https://developer.nvidia.com/transfer-learning-toolkit
Thanks for your quick response. Should i populate dataset with images in which contain both face and license plate in the same frame ? Or is it possible to any use any image that contains either face or license plate regardless of visual context for dataset ? It seems that first way results fewer images in my dataset but may improve generalization of model ?