Image retrieval example

[kloudkl]

This PR is meant to not distract #161. It does not compile yet and is blocked by #161.

[kloudkl]

The feature binarization step should be replaced by much more advanced algorithms[1][2].

[1] Minimal Loss Hashing for Compact Binary Codes, Mohammad Norouzi, David Fleet,
International Conference in Machine Learning (ICML), 2011.
[2] Yunchao Gong, Sanjiv Kumar, Henry A. Rowley, and Svetlana Lazebnik. Learning Binary Codes for High-Dimensional Data Using Bilinear Projections. In CVPR 2013.

[shelhamer]

Is this redundant given the merge of #161? If so, please close.

[kloudkl]

@shelhamer, no, this PR picked out image retrieval examples from #161. As you have commented, while feature extraction is of broad interests, image retrieval is only helpful to a few.

Large scale image retrieval cannot be performed using the real-valued features extracted from the fully-connected layers or the convolution layers. The mean-thresholding feature binarization method in 1915707 is of little practical value. So I'm working on a non-trivial algorithm in #274 which is planned to be finished in three days.

[OceanItAdmin]

Hi,
Can you please give some information on how to use image retrieval example... explanations in the code are note really explicit.
"This program takes in binarized features of query images and sample images"
" extracted by Caffe to retrieve similar images.\n"
"Usage: demo_retrieve_images sample_binary_features_binaryproto_file"
" query_binary_features_binaryproto_file save_retrieval_result_filename"
thanks,

[kloudkl]

@OceanItAdmin, a web demo written in Python will replace this one in a few days.

[shelhamer]

Replaced by #1028.