|Topic:||Metody redukce dimenze pro dataset funkční mapy světa|
|Department:||Vidění pro roboty a autonomní systémy|
|Supervisor:||Ing. Michal Reinštein, Ph.D.|
|Announce as:||Diplomová práce, Bakalářská práce, Semestrální projekt|
|Description:||The aim is to design and implement Deep Neural Network (DNN)[1, 2] based solution for the Functional Map of the World satellite imagery dataset (fMoW) and compare it with the state-of-the-art results achieved as part of the original challenge organized by IARPA. Due to complex nature of the fMoW dataset understanding the data is essential and therefore dimensionality reduction method UMAP should be implemented and used to explore and process the fMoW dataset. Comparison with related state-of-the-art in deep learning methods is integral part of the work. Instructions are as follows:
1. Explore the current state-of-the-art solutions of dimension reduction and deep learning algorithms, e.g. [1, 2].
2. Explore the fMoW dataset using the UMAP.
3. Design and implement a new solution to the fMoW dataset in Python using TensorFlow  framework.
4. Evaluate the solution and compare it to the state-of-the-art methods.
|Bibliography:|| Goodfellow, Ian, et al. „Deep Learning“, MIT Press, 2016
 He, Kaiming, et al. "Mask R-CNN" Proceedings of the IEEE international conference on computer vision. 2017.
 Christie, Gordon, et al. "Functional map of the world." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018. https://github.com/fMoW/dataset, https://www.iarpa.gov/challenges/fmow.html
 McInnes, Leland, John Healy, and James Melville. "UMAP: Uniform manifold approximation and projection for dimension reduction." arXiv preprint arXiv:1802.03426 (2018).
 Abadi, Martın, et al. "TensorFlow: Large-scale machine learning on heterogeneous systems, 2015." Software available from tensorflow.org.