COMP634 – Resit Assessment Deadline: 31st of July 2026

联系我们： 手动添加方式: 微信>添加朋友>企业微信联系人>13262280223 或者 QQ: 1483266981

COMP634 – Resit Assessment
Deadline: 31st of July 2026
With the rapid growth of online streaming platforms, the automated classification of
movie genres – typically performed using visual and/or textual information – has become
increasingly important for organizing and recommending movies and TV shows. In this
assignment, your aim is to build and train neural network models capable of classifying
movies into one of 9 genres, using images or text as input.
Specifically, you will use a small dataset composed of around 2500 movie instances
(image+text), which can be downloaded on Canvas. In this dataset, there is folder
(called “images/”) with all the movie posters (images), and a csv file (called
“dataset.csv”) in which each row has 3 columns: the first is the (movie poster) image
path, the second is the label/class, and the third is a text description (movie summary)
of the corresponding (movie) item. As mentioned before, there are 9 possible
classes/categories.
You will need to submit your code and a report. For the code, please use the provided
Jupyter notebook template, which is also available for download on Canvas. The
sections of this notebook match the sections your report should have, which are:
1. Data Management
a. In the Jupyter Notebook:
i. define an appropriate experimental protocol (such as k-fold, cross
validation, etc);
ii. create the dataloader to load the data; remember to include here
normalization, encoding, vocabulary creation, and other
techniques used to pre-process the data;
b. In the report:
i. detail the experimental protocol, i.e., how you split the data;
ii. explain the dataloader, including any pre-processing (such as
encoding, vocabulary creation, padding, normalization, etc)
performed in the input data.
2. Neural Networks
a. In the Jupyter Notebook:
i. create an efficient and robust Recurrent Neural Network (RNN)
(such as RNN, GRU, LSTM) to tackle the problem using only the
text data (last column of the .csv file); aim to make your
architecture as efficient (in terms of trainable parameters x
representation power) as possible;
ii. create an efficient and robust OR adapt and use an existing
Convolutional Neural Network (CNN) to tackle the problem using
only the image data; if you are proposing your own network, aim to
make your architecture as efficient (in terms of trainable
parameters x representation power) as possible;
iii. define the necessary components to train the networks (that is,
loss function, optimizers, etc);
iv. train the models;
v. for all training procedures, separately plot the loss and accuracy
with respect to the epoch/iteration.
b. In the report:
i. briefly describe your networks;
ii. explain your decisions regarding the components (loss function,
optimizers, etc);
iii. report table of the results (using appropriate metrics). Compare
the methods and define the best approach.
3. Evaluate models
a. In the Jupyter Notebook:
i. evaluate both models on the appropriate set.
b. In the report:
i. report results for the appropriate set using at least 3 different
metrics suitable for your problem. The selected metrics should
capture different aspects of model performance (avoid reporting
closely related measures);
ii. plot and discuss the confusion matrices.
Submission and Team Size
For this assignment, you must work independently – no group work!
The deadline is the 31st of July at 5PM.
You will need to submit your completed assessment on Canvas. Submit a single zip
file which includes a report (in pdf) and the python code (in .ipynb). Your report should
not be more than 1200 words long.
Your submission (that is, your .zip file) should have the name in the format of:
“studentID.zip” (example 201712345.zip).
Rubric
The project assessment against the following criteria:
1. Python code (60%)
a. Correct definition of the data split / protocol (5%)
b. Appropriate dataloader with correct encoding, vocabulary, padding,
normalization, etc (17%)
c. Correct definition of network architectures as well as necessary
components (18%)
d. Train the models and plot graphs (15%)
e. Test using the appropriate set (5%)
2. Report (40%)
a. Clearly and briefly describe all details of the experimental protocol (5%)
b. Explain the details of your dataloader, including any pre-processing
performed (such as encoding, vocabulary creation, padding,
normalization, etc) (12%)
c. Briefly describe the exploited network architectures, as well as the
components required for training (13%)
d. Report and discuss the training results table using appropriate metrics –
clearly define the best model and justify/discuss (5%)
e. Present and discuss a table and confusion matrices showing the final
results, using both models on the appropriate set and with suitable
metrics (5%)

COMP634 – Resit Assessment
Deadline: 31st of July 2026最先出现在KJESSAY历史案例。