Open JamesRick opened 4 years ago
I was able to train an svm on the same dev set used for ESC-10 and UrbanSound8K and make predicitons. Still need to move the code over to its own file and get a 20 iteration average like we have done for the affinity coding predictions.
Use an SVM on CNN/MaxPool layer outputs to get the supervised upper bound for each of the datasets.