How was the validation accuracy profile while training the foreground/background segmentation model with noisy segments obtained from the motion segmentation algorithm.

[aditya-vora]

Hi,
I just wanted to know, how was the validation accuracy profile when you were training a foreground/background segmentation model with AlexNet/CaffeNet architecture. From what accuracy did you start the training and what validation accuracy you obtained at the end. Were you getting low validation accuracy because of the noisy labels obtained because of the inaccuracy of the motion segmentation algorithm, or you observed a general trend of increasing validation accuracy?
Thanks,
Aditya Vora

[pathak22]

@aditya-vora The main trick in getting it work is the Trimap loss (described in the paper).

So, if you look at the validation loss on the full image without Trimap thresholding, it won't give you enough signal as the data is too noisy. But the Trimap loss on validation loss is a decent indicator (as it goes down) -- so tracking it is useful.