ssd_keras
ssd_keras copied to clipboard
implementing prior bounding boxes as keras layers?
is there any particular reason to implement the bounding box priors as a layer in the keras model? if i'm reading this correctly, it only ever gets used again during post-processing of the prediction. is it to 'auto-configure' the priors based on input size etc... by detecting input from previous layer?
j
@BenMacKenzie yep, it's for auto-configuring priors based on a layers' outputs.
Make PriorBox a keras layer, will it compute input gradient in backward?