https://github.com/neild0/siaception
Development of few-shot learning tool for a variety of image classes, utitilizing features extracted from ConvNets.
https://github.com/neild0/siaception
convolutional-neural-networks few-shot-learning keras ml nns one-shot-learning pretrained-models siamese-neural-network
Last synced: 7 months ago
JSON representation
Development of few-shot learning tool for a variety of image classes, utitilizing features extracted from ConvNets.
- Host: GitHub
- URL: https://github.com/neild0/siaception
- Owner: neild0
- Created: 2018-12-29T05:53:22.000Z (almost 7 years ago)
- Default Branch: master
- Last Pushed: 2018-12-29T19:34:01.000Z (almost 7 years ago)
- Last Synced: 2025-01-22T05:28:47.523Z (9 months ago)
- Topics: convolutional-neural-networks, few-shot-learning, keras, ml, nns, one-shot-learning, pretrained-models, siamese-neural-network
- Language: Jupyter Notebook
- Homepage:
- Size: 5.04 MB
- Stars: 2
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# SiaCeption - Siamese Networks Based on Pretrained ConvNets
Development of few-shot learning tool for a variety of image classes, utitilizing features extracted from ConvNets. This is specifically for my school Comenius project.
A big problem currently in the field of ML is the fact that a lot of data is required to train image classification neural networks. By using a siamese network approach in this project, I hope to make algorithms that require (as shown in the test) only a single image (although a few more do increase the accuracy) to train.
## Follow the attached jupyter notebook to understand the algorithm and go through an example on a bunch of animals. (Developed with only 1 image for training each class =)