Ai Assignment Report
-
Upload
raghav-iyer -
Category
Documents
-
view
228 -
download
0
Transcript of Ai Assignment Report
-
7/30/2019 Ai Assignment Report
1/20
AI ASSIGNMENT REPORT
Ayush Agarwal
201001107
Q1
Dimension = 20
-
7/30/2019 Ai Assignment Report
2/20
Dimension = 10
Dimension = 2
As the dimension increase the accuracy increases and the recognazibility also increases.
Dimension Acuuracy
2 30.8%
-
7/30/2019 Ai Assignment Report
3/20
10 95.8%
20 99.5%
As we can observe that when we take 2 dimensions the accuracy is very low and when the dimension
equal to 10 then the accuracy increases by huge amount and after this the accuracy increases very less.
It saturate because eigen vector are sorted by their values (weight of vectors) which contributes to the
feature vectors thats why the accuracy after top 10 vectors increase very less.
Q2 : MEAN FACE
-
7/30/2019 Ai Assignment Report
4/20
-
7/30/2019 Ai Assignment Report
5/20
Q3 :A & B SAMPLE IMAGE 1
-
7/30/2019 Ai Assignment Report
6/20
The original image is from input dataset .
Here we can observe that as the d value increases the RMSE value decreases and recognizability also
increases because increasing d mean we are using more eigen vectors to reconstruct our image
therefore the eigen vectors which have less eigen value or weight also included in reconstruction
therefore the information loss in reconstruction is very less which is shown by decreasing value of RMSE.
-
7/30/2019 Ai Assignment Report
7/20
As the d increases the RMSE decreases because as the information for reconstruction increases the error
in reconstructing image also decreases.
-
7/30/2019 Ai Assignment Report
8/20
SAMPLE IMAGE 2
-
7/30/2019 Ai Assignment Report
9/20
-
7/30/2019 Ai Assignment Report
10/20
-
7/30/2019 Ai Assignment Report
11/20
This image is my face image which is present in the dataset.
We can observe that the same as in above sample image but in this sample the recognizibility is very
less because of the less illumination and some unwanted region ( background) in the input image and
the other which are present in the dataset.
-
7/30/2019 Ai Assignment Report
12/20
SAMPLE IMAGE 3
-
7/30/2019 Ai Assignment Report
13/20
-
7/30/2019 Ai Assignment Report
14/20
This image is not present in the input dataset.
Here we can observe same relation between the d and RMSE value as in the above two sample images.
But in the above two sample the RMSE values are lower than as compare to this sample because image
of this person is not present in the input dataset. Hence the recognazibility of the person is also low as it
reconstructing the nearest image or which have minimum distance from the feature vector of the givenimage.
Inserted only three samples from the each sample image , the other image form d 1 to 15 is present in
the folder ( sample1, sample2 ,sample3)
C.
Input Image
-
7/30/2019 Ai Assignment Report
15/20
-
7/30/2019 Ai Assignment Report
16/20
We can observe that as the masking image size increases the reconstructed image recognazibility
decreases because of the loss of features of the input image.
It reconstruct the feature vector in to image which is nearest to the input image feature vector in
feature space.
D.
-
7/30/2019 Ai Assignment Report
17/20
In this case the input image is of tree but we have trained our system using faces only therefore when
we extract features from the input image and represent it in low dimension feature space it calculate
the feature vector in the space and reconstruct the image. Therefore the reconstructed image is a face.
The difference in the reconstructed image of this from the above one Is because of the features (we can
see that in above input image the mean color intensity is low but in this image the mean color intensity is
high therefore the reconstructed face is bright as compare to the above reconstructed face).
-
7/30/2019 Ai Assignment Report
18/20
In all the above sample we can observe that all the object are reconstructed into face image which is
obvious because we trained our system with face images and in our feature space all the features vector
correspond to the face image, and when we reconstruct the image from the feature vector of any input
image it somehow give the mean face or near to the mean face.
E.
-
7/30/2019 Ai Assignment Report
19/20
We trained the system using only one image of the person without spectacles that why the
recognazibility of the reconstructed image is very less and the RMSE is very high.
-
7/30/2019 Ai Assignment Report
20/20
We trained the system using only one image of the person without spectacles that why the
recognazibility of the reconstructed image is very less and the RMSE is very high.
We can also observe that because of the sunglasses the face reconstructe have low intensity in the eye
region as compare to above image.