Few Shot Learning is the next generation class of algorithms being researched currently for object categorization tasks in Computer vision with unseen data and without pre-training.
Some popular algorithms:
- 'Matching networks for one shot learning'
- 'Learning to compare: Relation network for few-shot learning'
- Prototypical Networks algorithm for few shot learning
This repo demonstrates experiments conducted with the prototypical networks algorithm as published in this paper. Two key changes improves the accuracy (88%) of a 5-way 5-shot classification task, the details of which are provided in the report.