Object Detection with Bootstrapping Carlos Rubiano Mentor: Oliver Nina
Began Testing with Multiple Networks Trained 3 Networks with 33%, 50%, 100% randomized dataset Using the trained models from Cifar10 - Tested on same test batch with different networks to obtain predictions Applied different techniques for voting - Voting with labels - Highest probability from 3 networks
Results Possible randomization issue: networks ran with re- sampling score lower Voting did increase accuracy
Used original data for highest scores Tried another voting technique - Adding, averaging, and picking highest Tested with more than 3 networks
Modified code to improve runtime speed - Allows getting probabilities for multiple networks much faster by implementing it into a function and then multi-threading with a bash script