This assignment is the machine learning project for classifying images. You will have to train a machine learning model to predict the $20$ possible classes of images. The

training set contains these labels $($along with the number of images in each class$)$:

Class ID Class Name Count

$0$ book $739$

$1$ bottle $715$

$2$ car $500$

$3$ cat $564$

$4$ chair $508$

$5$ computermouse $534$

$6$ cup $774$

$7$ dog $464$

$8$ flower $875$

$9$ fork $605$

$10$ glass $572$

$11$ glasses $436$

$12$ headphones $459$

$13$ knife $790$

$14$ laptop $446$

$15$ pen $954$

$16$ plate $433$

$17$ shoes $899$

$18$ spoon $625$

$19$ tree $591$ A subset of the images collected in Assignment $1,$ Exercise $1(12483$ images$)$ is provided for training

$($download$),$ the remaining images serve as hidden test set.

Note that the images have been validated and thus have numbered file names, where the labels are

stored separately in the accompanying labels.csv file. Furthermore, the images have been resized

to $100$ pixels $($either width or height, depending on the aspect ratio$).$ In addition to the challenge server submission, you must also upload all of your project files to

Moodle. This will be used to check for plagiarism and to verify that your model corresponds to the

uploaded submissions.

Create a ZIP archive that includes all project files, i$.$e$.,$ the entire source code, all configuration files

and all documentation files. The filename of the ZIP archive can be arbitrary. You do not have

to include the training data itself, however any scripts that augment the training data need to of

course be included so that we can exactly replicate your result.

General Project Hints:

$$ Divide the project into subtasks and check your program regularly.

$$ Decide which samples you want to use in your training, validation or test sets.

$$ Check the units on data loading, neural network inference and training and the respective

code files and decide which elements are useful for your implementation.

$$ Implement the computation of the loss between NN output and target $($see code files of

Unit $7).$

$$ Implement the NN training loop $($see code files of Unit $7),$

$$ Implement the evaluation of the model performance on a validation set $($see code files of

Unit $7).$

$$ You can have a look at the example project in the code files of Unit $7.$ This is not a classification

project, but you can still check whether you can use parts of the code as well.

$$ Set a random number seed in order to obtain reproducible results.

$$ Do not use transfer learning, you should create a model from scratch. You can of course

research published models, but do not copy $1$:$1.$

$$ You only have $10$ attempts to submit predictions, so it will be important for you to use some

samples for a validation set and maybe another test set to get an estimate for the generalization

of your model.

$$ You do not need to reinvent the wheel. Most of this project can be solved by reusing parts of

the code materials and assignments from this semester.

$$ When uploading to the challenge server, you might not immediately see a result due to the

server$$s task scheduling. Please be patient and check back later. This also means that you

should try to upload your model in due time to avoid having no immediate feedback when the

assignment deadline approaches.