Code overview

functions:

• convertAllImages.m: convert all images from JPG format to PNG format.
• extractAllImages.m: extract features from all images, using 'compute_descriptors' library in lib/feature_detector.
• quantizeAllImages.m: apply flann kd-tree to get the nearest bins
• makeAllBow.m: make BoW vector by assigning features to the the nearest bins and computing their weights

groundtruth:

• compute_ap.cpp: compute average precision for each query

lib:

feature_detector:

• compute_descriptors.exe, compute_descriptors.mac, compute_descriptors_32bit.ln, compute_descriptors_64bit.ln: extract features and compute descriptors for Win/MAC/Linux32/Linux64, respectively

flann-1.8.4: fast library for approximate nearest neighbors, is used to compute neaest clusters for each feature (http://www.cs.ubc.ca/research/flann/)

vlfeat-0.9.19: library of computer vision algorithms, some minor functions of the library are used (http://www.vlfeat.org/)

Steps:

• createParams.m: store parameters, data locations
• step1_extract.m: extract features of all query and data images
• step2_quantize.m: perform quantization
• step3_query.m: process all queries and retrieve ranked lists
• step4_evaluate.m: compute mean average precision of all ranked lists

How to run

• compile groundtruth/compute_ap.cpp to get the suitable execution file for your os.
• copy the data to directory: data/image/
• copy the queries to directory: query/image/
• copy groundtruth files to directory: groundtruth/
• run compile_script to initialize the groundtruth program (compile_script.bat for Windows, compile_script.sh for Linux & Mac).
• run step1, step2, step3, step4 sequentially.
• ranked list are outputted to directory: ranklist/
• average precision (AP) of each query is written to the directory: ap/
• mean average precision (mAP) is printed on the screen after running script step4.