https://github.com/agrover112/siamesenet-search
MNIST digits image similarity search by Indexing with Annoy and using trained embeddings from a Siamese Net with Triplet Loss .
https://github.com/agrover112/siamesenet-search
annoy image-similarity indexing mnist siamese-architecture siamese-network siamese-neural-network similarity-search tensorflow-examples tensorflow2
Last synced: 8 months ago
JSON representation
MNIST digits image similarity search by Indexing with Annoy and using trained embeddings from a Siamese Net with Triplet Loss .
- Host: GitHub
- URL: https://github.com/agrover112/siamesenet-search
- Owner: Agrover112
- License: mit
- Created: 2020-11-25T23:57:24.000Z (almost 5 years ago)
- Default Branch: main
- Last Pushed: 2022-02-13T16:11:45.000Z (over 3 years ago)
- Last Synced: 2025-02-07T20:23:42.218Z (8 months ago)
- Topics: annoy, image-similarity, indexing, mnist, siamese-architecture, siamese-network, siamese-neural-network, similarity-search, tensorflow-examples, tensorflow2
- Language: Jupyter Notebook
- Homepage:
- Size: 617 KB
- Stars: 5
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# SiameseNet-Search
Search can be performed in various ways and using various architectures. In the following repository embeddings are created by using an non-cnn **base embedding model** which is tuned using a *Siamese Neural Net* with *Triplet Loss* inspired from [F. Schroff, D. Kalenichenko and J. Philbin, "FaceNet: A unified embedding for face recognition and clustering," 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA, 2015, pp. 815-823. ](https://arxiv.org/pdf/1503.03832.pdf) paper from Google , on **10K MNIST digits** images from train-set.The trained model is used to create embeddings for *entire* train-set and then Indexed using [Approx Nearest Neighbours](https://github.com/spotify/annoy) being faster than K-NN.
More like a prototype.### Embedding Space Visualized
PCA used for visualizing the embedding space of **10K images** of different classes.

## References
- https://arxiv.org/pdf/1503.03832.pdf
- http://www.cs.cmu.edu/~rsalakhu/papers/oneshot1.pdf
- http://yann.lecun.com/exdb/mnist/
- http://cseweb.ucsd.edu/~dasgupta/papers/rptree-stoc.pdf
- https://github.com/spotify/annoy
- https://github.com/PotatoSpudowski/S.I.D.E