Captioned image retrieval

We explore the task of retrieving similar captioned images from a dataset, given a previously unseen captioned image.

Note: the source code in SpatialPyramid has some bugs fixed. It is not exactly the same as the original source code from UIUC.

To run baseline

1. Add SpatialPyramid to path
2. Unzip the Flickr 8k dataset to a data subdirectory
3. Run close all; clear all; baseline;

To run crawling

1. Add the required search tags into searchTags array in line 13 of crawler/gallerySearch.py
2. Execute python crawler/gallerySearch.py

Note: The crawled Imgur dataset (DataM) consists of 32K images and ~110K captions and is sized at ~16 gb. It can be provided on request.

To run LDA

See readme in lda/

To run CNN

See readme in caffenet/

DataM dataset

The dataset and associated captions can be found at: http://pages.cs.wisc.edu/~ms/CS766-ComputerVision/captioned-image-retrieval/