|
Hi, I am working on image classification. It seems to me that Fisher vector based on Gaussian Mixture Models outperforms other approaches such as Sparse Coding and Bag of visual words on various datasets. But the dimensionality of Fisher Vectors are very high. I am trying to implement a discriminative fisher vector for (1) improving the fisher vector's performance (2) getting a better performance with a reduced number of features compared to Fisher Vector (FV). I found only one paper for discriminative fisher vector (DFV) 'http://researchweb.iiit.ac.in/~siddhartha.chandra/ICVGIP2012-Final.pdf' link text. I implemented it using vlfeat library but I am unable to reproduce the results. In my case I am getting very similar performance using DFV compared to FV. I would like to know: (1) any relavent literature on discriminative fisher vector (2) some directions or ideas for getting a better performance using discriminative fisher vector |
