Code for Explain to Fix: A Framework to Interpret and Correct DNN Object Detector Predictions paper presented at NIPS 2018 Workshop on Systems for ML
Merge code into original Caffe/SSD branch and recompile. Source code modifications change output detection format to include detection indices. All other processing is done by Python scripts in examples/e2x folders. Details for each script is given by "--help" command. Current script support SSD300 model as in the paper. Other models and datasets can be easily added.
