Classification of Household Materials via Spectroscopy
Z. Erickson, N. Luskey, S. Chernova, and C. C. Kemp, "Classification of Household Materials via Spectroscopy", arXiv, 2018.
Project webpage: http://healthcare-robotics.com/smm50
Download the SMM50 dataset
Use the following commands to download and extract the SMM50 dataset.
cd data wget -O smm50.tar.gz https://goo.gl/2X276V tar -xvzf smm50.tar.gz rm smm50.tar.gz
Running the code
Our residual and vanilla neural networks are implemented in Keras with the Tensorflow backend.
Results presented in tables I and II from the paper can be computed using the following.
python learn.py -t 0 -a svm python learn.py -t 0 -a nn python learn.py -t 0 -a residualnn
Generalization with leave-one-object-out validation results from table III can be computed using the commands below. These results are also used for Fig. 12 and 13 in the paper.
python learn.py -t 1 -a svm python learn.py -t 1 -a nn python learn.py -t 1 -a residualnn
The generalization results with increasing numbers of obects can be recomputed using the commands below. This corresponds to Fig. 15 in the paper.
python learn.py -t 2 -a residualnn
All of the plots from the paper can be regenerated using
plot.py. This requires plotly.