Hello everyone. I hope you are doing well.
I’m looking forward to learn more about the inner workings of the DCF Tracker provided by the VPI API. One of the things I came across while working on Object Tracking is the Model Masking Window and Feature Masking Window parameters of the vpiSubmitDCFTrackerUpdateBatch function.
My goal is to try implement the Spatial Reliability Map that is done in the CSR-DCF paper, from which the VPI DCF Tracker is based on. I would then use that map in the UPDATE phase of the algorithm to filter out the pixels I don’t want the tracker to learn. I’ve thought about having a segmentation phase first to get the object mask and then feed it to the algorithm.
I suppose that my mask would fit in the Model Masking Window parameter instead of the Feature Masking Window, but I don’t quite understand why both parameters have the same dimensions. Could someone point me to the difference between these two windows? Also, would this approach be feasable or does this tracker not support a Spatial Reliability Map-like mask?
Thank you in advance. Best regards.
jamorim