encode Fisher

[20155004]

i think that rid2desc map to the fisher vector. so, fisher = vl_fisher(rid2desc,gmm.mean ~~);

how can i use other inputs such as rid2tlbr, imsize,spatailPyramid ??

[Dong-JinKim]

20153080

you need to encode multiple fisher vectors with spatial pyramid in 1131, which means encode from 1_1(original size) region and from 3 3_1 divided regions. so totally encode 4 vectors. what you did was

fisher = vl_fisher(rid2desc,gmm.mean ~~);

was only for 1*1 full region. you now need to encode fisher vectors from divided regions. in order to do that, you need to select activation descriptors corresponding to certain region. that's why you need to use rid2tlbr and imsize

[20155004]

thank you, we have rid2desc matrix (N x331)

then why the fisher = vl_fisher(rid2desc,gmm.mean ~~); was only for 1 region??

rid2desc already includes descriptors of all divided regions..am i wrong??

that's confusing me..

[Yoon-Jaeshine]

20153395 Yoon Jaeshin

It means that,,

1. fisher1=vl_fisher(rid2desc,gmm.mean~)
2. fisher2=vl_fisher(a,gmm.maen~)
3. fisher3=vl_fisher(b,gmm.mean~)
4. fisher4=vl_fisher(c,gmm.mean~)

rid2desc are all descripotrs. a are descriptors from just upper part of images in each scale. b are descriptors from just middle part of images in each scale. c are descriptors from just lower part of images in each scale.