Need smaller SVM

[pfabreu]

I've been trying to run this but I ran out of GPU memory. Therefore I reduced the batch size and it works fine (i.e no cuda crash) however now I'm facing the problem that your SVM does not take batch reduction into account. Is there any workaround to this (or would you be so kind as to train the SVM again just with less features, I would be eternally grateful) or my only chance is really to get a better GPU?

EDIT: as a botched solution I just did:

    % load the trained SVM
    SVM = load('data/rcnn_models/DeepPed/SVM_finetuned_alexnet.mat');
    %PersonW = SVM.W; %Feature weights for scoring
    PersonW = SVM.W(1:end/2);
    PersonB = SVM.b; %Constant scoring factor

But I'm pretty sure this will be incorrect.

Thank you