experimental setting

train snn loss and eval f1-scores

added f1 scores

working model with z-scores eval and softmax formulation

Experiments: changes from old commit

feat: added 3dSplineCNN and MLP for learning z-depth coordinate

fix: better logging of K graphs for training

feat: training and evaluating on K graphs

feat: 3dSpline defined in NMT (not used) + test script

working version of snn-loss performance equivalent to infonce-loss

added API for mgmsolver with logit and other funcs for unary costs, added cost margin in loss and added gatedMLP for keypoints features

feat: mgmsolver using official api

added GenericInfoNCE loss

added hyperspherical loss for SNN-Loss and mgmsolver function for unary costs

feat: added mgm solver in inference and testing scripts