Has anyone tryed to mess with the depth_multiplier variable,in the config files, during a training with Google's Object Detection API?? In my understanding if you change the value of this variable (1.0 by default) you can't make use of fine tuning from the already provided pre-trained models. Am i correct? Thanks in advance.
yes, you are correct... if you were to do this, you'd have to train from scratch, or first train an Imagenet classifier with the same base architecture and fine tune from that.