baucheng / itq-image-retrieval Goto Github PK

=======================================================================================

Overview:

This matlab source code comes from http://www.unc.edu/~yunchao/itq.htm.

I have modified it and implement drawing the Precision Recall Curves using
different methods such as ITQ/LSH/RR/SKLSH, and different binary bit length
PR_Curves using LSH. A Retri_Images_Demo is also provided to show the image
retrieval result using ITQ.

Modified by Willard, and you can see the detains from my website is:

www.yuanyong.org/blog/cv/lsh-itq-sklsh-compliment

========================================================================================

Important files are:

1. ITQ.m: implements the Iterative Quantization method
   

2. compressITQ: convert large-scale data to binary matrix
   
3. main.m: demo code computes recall precision, also implements some other baseline methods
   
4. RF_train.m: estimate some parameter for SKLSH
   
5. RF_compress.m: perform SKLSH binary embedding
   
6. cca.m performs a canonical correlation analysis for supervised cases
   
7. Demo_PR.m: This is a geometric illustration of Draw the Recall Precision Curve
   
8. DemoLSH_PR.m: This is a geometric illustration of LSH recall and precision using
   different code length
   
9. Retri_Images_Demo.m: This is a PCA-ITQ demo showing the retrieval sample
   

========================================================================================

Some important notes:

1. before ITQ, the data must have be centered
   

2. we found 20-50 iterations are usually enough for ITQ
   
3. I modified it on a 16G memory computer, 4G is failed for the dataset is large
   

========================================================================================

Demo results

Note: The bounding box image using red color is the query image.

If there is a bug, don't hesitate to contact me. Thanks!