Multimodal TV Show Segmentation

I will continue from last year’s work and improve the clustering algorithm to the in-production code and enhance the previous work. The main problem is to find the correct anchor. For this, I want to propose a celeb detection API of Microsoft azure which is the best for this use case. Currently, the most time-consuming process in the program is going frame by frame and extracting faces. The face-recognition method used in the production code processes each frame individually. I want to upgrade it to a parallelized algorithm to process multiple frames in a batch and increase the processing speed exponentially, which will also help in faster testing of hyperparameters. Batch processing can be much quicker than processing single images at a time. So I think we use batch processing for multi-threading.