Disaster Image Retrieval from Social Media [DIRSM] -- Multimedia-Satellite-Task 2017
This code was conceived for the Disaster Image Retrieval from Social Media [DIRSM] subtask, part of the Multimedia-Satellite-Task (MediaEval Benchmark).
It allows fine-tuning of several ConvNets using TensorFlow framework. Among the networks, it includes:
- AlexNet
- VGG16
- GoogleNet
- ResNets(50,101,152) with bootleneck
- DenseNets(121,169,201,161) with bootleneck and compression
DenseNet models were load using python-torchfile, ResNets were created using the models available at tensorflow-resnet and, finally, other models were converted to npy with caffe-tensorflow and then loaded.
The code also include algorithms for ranking generation and evaluation (Mean Average Precision).
If you use this code in your research, please consider citing:
@inproceedings{multiBrasil_mediaeval,
author = {Nogueira, Keiller and Fadel, Samuel G. and Dourado, \'{I}caro C. and Werneck, Rafael de O. and Mu\~{n}oz, Javier A. V. and Penatti, Ot\'{a}vio A. B. and Calumby, Rodrigo T. and Li, Lin T. and dos Santos, Jefersson A. and Torres, Ricardo da S.},
title = {Data-Driven Flood Detection using Neural Networks},
year = {2017},
booktitle=medeval,
location = {Dublin, Ireland},
url={http://slim-sig.irisa.fr/me17/Mediaeval_2017_paper_39.pdf},
pages={2}
}