Performed both within the same branch because I needed similar scripts for both for visualizing GNN/ADAM performance between different experiments.
There are 4 important features added here:
Script to generate confusion matrix for GNN outputs given an input curriculum
Script to generate confusion matrix for ADAM outputs given an input curriculum
Script to generate confusion matrix for top GNN outputs per object given an input curriculum
Added functionality to
Retrieve top k GNN decodes per object from shape_stroke_graph_inference.py
Add each of the top k decodes as a node in ADAM and handle accordingly.
Aside from those, most of the new additions here are just adding parameter files which I used the preexisting downsample/unknown object generation scripts to create the curricula for.
Closes #1156 and #1157
Performed both within the same branch because I needed similar scripts for both for visualizing GNN/ADAM performance between different experiments.
There are 4 important features added here:
kGNN decodes per object fromshape_stroke_graph_inference.pykdecodes as a node in ADAM and handle accordingly.Aside from those, most of the new additions here are just adding parameter files which I used the preexisting downsample/unknown object generation scripts to create the curricula for.