Meeting minutes for 31 Aug 2016: Tom and Luis
=====================
What I have been doing:
– I have been working on the optimisation of the CLR boundaries. Particularly on the optimisation of the base_lr and max_lr.
Now the train code takes two extreme values 0.01 and 10e-15, and moves linearly from the base_lr to the max_lr (as
recommended by the CLR paper) saving a snapshot of the network after a certain amount (depends on free space in the SDD)
of steps.
– (After meeting) I have sent an email to the organisers of the CARE workshop asking the length of my presentation, in case that
it is a presentation, which is still unknown as the program has not been published in the workshop website yet.
Tasks to do:
– Send an email to Dominique as soon as I know the length of my presentation.
– Remember to write in the journal paper that as the upsampling layers are modified because we use a different number of classes
than the original network, the upsample kernels are initialised to bilinear interpolation.
– The journal paper has priority over the detection challenge for MICCAI, but try to have something quick and dirty for the challenge.
– First draft of the paper before going on vacations.