It should be implemented using Matlab. Please help! Don't need to be perfect, I just need something at least make sense to turn in. Thank you :)

[2] For each image compute its dense SIFT and use Fisher Vector to aggregate. Training SIFT data set far PCA and GMM computation is provided as belaw, with the training SIFT data at: https://umkc.box.com/s/3shyqe1mkvb6n19arnrusdastwqms3rs [20ptsl at:1 % train SIFT PCA, gmm doTrainGMM 0 if doTrainGMM % load SIFT training data load data/hw-2-data.mat; [A, s, latlzprincomp double(sift gmm_km_train)): fiqure(21): grid on; hold on; plot(lat, -: 1] Use pooled (2x2) color histogram to represent images provide the two following functions for feature extraction and distance computing (codebook can be just fixed 8x4x2 HSV uniform quantization) [20pts kda116, 24l ne 16, 32, 64, 96: t0-cputime; for j-1Henath(kd) function [hl-gctPagledHSVHi istogram(im, codebaok, pooling) function [distJ-getPooledHSVDistancelim1, im2) for k-:lenathinc) 2] Compute HoG feature for image, use the average, as well as 2x2 pooled average as texture feature. Use block size of 8 pixeland 2x2 cell structure. (Hint: use reb2gral.bl) 20pts] function [hogl-getimHoslin teriottan t=%1.2t GM M: kd(%d)-%d, nc(%d-%d: ,, cputime-to, j.kd(j), k, nc(k) unction [hogl etHoRisth1, h2) % training data reduce dimension via PCA x double(sift gmm_km trainxA Lkd): % gmm model: lgrum(lds).rn, gmm(45)ER9mm(j' k).pl=v1.9mrn(x, S(k). MaxNumiterations, 30) hog dist dense sift fv dist end 20 20 40 60 80 100 % save PCA and GMM model: 100 save data/hw2-A-gmm-kd-16-24-nc-16-32-64-96.mat A gmm; 20 40 60 80 100 20 40 60 80 100 sift distance gnd truth 20 load data/hw2-A-gmm-kd-16-24-nc-16-32-64-96.mat A gmm; 40 end 60 60 80 100 create a functions: 100 function Id-cetirExisift, A. Fmm) 20 40 60BO 100 20 40 60 80 100 S] Fuse the distances from different features, and try your own way af finding the best mixing parameters, Le For images and their Color Histogram, HeS, and FV aggregated dense SIFT features respectively. Justify your choice of welghts, and plot TPR-FPR ROC curves for different cholces. [20pts] [4] For the 2x2 pooled HSV feature, Hog feature, Fisher Vector aggregated dense SIFT feature please compute the nx n distance map between all image pairs, and plot their TPR-FPR separately ht: rog). 20pts. Examples of distance map: Example: how it should look like hog dist gnd truth 300 0 hist dist hog dist sift distance 2 1 dense sift fv dist 20 fused dist ROC AUC: 09.56% EER 45.56%) ROC CAUC: 26.78%, EER 70.00%) ROC AC 51.56% EER: 51.11%) ROC (AUC: 6a67%, EER 30.00%) ROC A C 69.89%, EER 30.00%) false positive rate false positive rate talse positive rate false positive rate false positive rate [2] For each image compute its dense SIFT and use Fisher Vector to aggregate. Training SIFT data set far PCA and GMM computation is provided as belaw, with the training SIFT data at: https://umkc.box.com/s/3shyqe1mkvb6n19arnrusdastwqms3rs [20ptsl at:1 % train SIFT PCA, gmm doTrainGMM 0 if doTrainGMM % load SIFT training data load data/hw-2-data.mat; [A, s, latlzprincomp double(sift gmm_km_train)): fiqure(21): grid on; hold on; plot(lat, -: 1] Use pooled (2x2) color histogram to represent images provide the two following functions for feature extraction and distance computing (codebook can be just fixed 8x4x2 HSV uniform quantization) [20pts kda116, 24l ne 16, 32, 64, 96: t0-cputime; for j-1Henath(kd) function [hl-gctPagledHSVHi istogram(im, codebaok, pooling) function [distJ-getPooledHSVDistancelim1, im2) for k-:lenathinc) 2] Compute HoG feature for image, use the average, as well as 2x2 pooled average as texture feature. Use block size of 8 pixeland 2x2 cell structure. (Hint: use reb2gral.bl) 20pts] function [hogl-getimHoslin teriottan t=%1.2t GM M: kd(%d)-%d, nc(%d-%d: ,, cputime-to, j.kd(j), k, nc(k) unction [hogl etHoRisth1, h2) % training data reduce dimension via PCA x double(sift gmm_km trainxA Lkd): % gmm model: lgrum(lds).rn, gmm(45)ER9mm(j' k).pl=v1.9mrn(x, S(k). MaxNumiterations, 30) hog dist dense sift fv dist end 20 20 40 60 80 100 % save PCA and GMM model: 100 save data/hw2-A-gmm-kd-16-24-nc-16-32-64-96.mat A gmm; 20 40 60 80 100 20 40 60 80 100 sift distance gnd truth 20 load data/hw2-A-gmm-kd-16-24-nc-16-32-64-96.mat A gmm; 40 end 60 60 80 100 create a functions: 100 function Id-cetirExisift, A. Fmm) 20 40 60BO 100 20 40 60 80 100 S] Fuse the distances from different features, and try your own way af finding the best mixing parameters, Le For images and their Color Histogram, HeS, and FV aggregated dense SIFT features respectively. Justify your choice of welghts, and plot TPR-FPR ROC curves for different cholces. [20pts] [4] For the 2x2 pooled HSV feature, Hog feature, Fisher Vector aggregated dense SIFT feature please compute the nx n distance map between all image pairs, and plot their TPR-FPR separately ht: rog). 20pts. Examples of distance map: Example: how it should look like hog dist gnd truth 300 0 hist dist hog dist sift distance 2 1 dense sift fv dist 20 fused dist ROC AUC: 09.56% EER 45.56%) ROC CAUC: 26.78%, EER 70.00%) ROC AC 51.56% EER: 51.11%) ROC (AUC: 6a67%, EER 30.00%) ROC A C 69.89%, EER 30.00%) false positive rate false positive rate talse positive rate false positive rate false positive rate