Assignment 8 – CMPUT 328
Semantic Segmentation
Semantic Segmentation on Texture Images dataset:
Introduction: The Texture Images dataset was created by placing these 4 texture images
randomly on blank images of size 196x196:

A sample image by this procedure is shown below:

The dataset can be found inside the “TextureImagesDataset” folder in your code directory. There
are 2000 samples in the train set and 500 samples in the test set.
Each sample consists of an RGB image of size (1, 196, 196, 3) and a mask of size (1, 196, 196, 1).
They are both of uint8 type. The mask tells us which texture each pixel belongs to. It will have
value from 0 to 4: 0 is the background class (i.e. that pixel belong to no class), 1-4 are for the 4
classes of the 4 texture.
The whole train set will have shape (2000, 196, 196, 3) for the images and (2000, 196, 196, 1) for
the mask. Similarly, test set will have shape (500, 196, 196, 3) for the images and (500, 196, 196,
1) for the mask.

Task: Your task in this question is to do semantic segmentation on this dataset. To be specific, you
just have to write the structure of the semantic segmentation network in the function SemSeg().
Everything else has been done for you. Remember, the returned output for this function must
have size (, 5, 196, 196). There are no architecture requirements or guidelines.
MARKING:
25% Marks: Online lab quiz will be performed during (Monday, Nov 4 – Friday, Nov 8). Note
that you must have to be present in your designated lab section. You will not get any mark for
this part if you don’t attend in your own lab section.

75% Marks: The due date is Friday, November 8 by 11:55 pm. Your mark will scale
linearly with the pixel accuracy of your segmentation, starting from 90.5% (0 mark) and cap
at 98.5% (75 Marks)

SUBMISSION:
You need to submit the complete solution of file: ‘Assignment_8_StudentID.ipynb’. Also, the
model file ‘model.pt’ as well. Write down your student name and student id at the top of the file.
Keep the output block of the final section while submitting your solution. The output block
may look like this: