Abstract | ||
---|---|---|
This paper tackles the unsolved important problem of training deep models with small amounts of annotated data. We propose a semi-supervised self-training bootstrap to deep learning on small datasets by retrieving and utilizing additional images from internet image search. We adapt the pseudolabel method proposed by Dong-Hyun Lee in 2013, previously used on the elementary MNIST handwritten digit classification task. We show that by suitable modifications to its example weighting and selection mechanisms it can be adapted to general image classification tasks supported by online image search. The proposed approach does not require any human supervision, it is practical and efficient, and it actively avoids overtraining. The usefulness of the proposed method is demonstrated on the SUN 397 dataset with only 50 training images per category. When exploiting results of Google's Image Search, we achieve a significant improvement, with a classification accuracy of 51%, as opposed to 39% without these results. |
Year | DOI | Venue |
---|---|---|
2016 | 10.1145/2948628.2948633 | SCCG |
Keywords | Field | DocType |
convolutional neural network, deep learning, image classification, reinforcement learning | Weighting,MNIST database,Convolutional neural network,Computer science,Artificial intelligence,Deep learning,Contextual image classification,Bootstrapping (electronics),The Internet,Reinforcement learning,Computer vision,Pattern recognition,Machine learning | Conference |
Citations | PageRank | References |
0 | 0.34 | 6 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Martin Kolár | 1 | 0 | 0.34 |
Michal Hradis | 2 | 132 | 14.19 |
Pavel Zemcík | 3 | 120 | 24.73 |