OPTIMIZATION: Image Classification, Semantic Segmentation. Adversarial Attack
Budget: €30 – €250 EUR
Python. Kaggle
This assignment consists of three main parts for which we expect you to provide code and
extensive documentation in the notebook:
• Image classification
• Semantic segmentation
• Adversarial attacks
In the first part, you will train an end-to-end neural network for image classification. In
the second part, you will do the same for semantic segmentation. via the Kaggle competition. In the third part, you will try to find and exploit the weaknesses of your classification and/or segmentation network. For the latter there is no competition format, but we do expect you to put significant effort in achieving
good performance on the self-posed goal for that part. Finally, we ask you to reflect and
produce an overall discussion with links to the lectures and “real world” computer vision.
In general, we will evaluate the correctness of your approach and your understanding of what you have done
that you demonstrate in the descriptions and discussions in the final notebook.
!!! : The codes are basically done, performing a really poor accuracy. The intended goal is to make optimization over them and specify the changes and the reasons.
The adversarial attack has errors. does not run.
This assignment consists of three main parts for which we expect you to provide code and
extensive documentation in the notebook:
• Image classification
• Semantic segmentation
• Adversarial attacks
In the first part, you will train an end-to-end neural network for image classification. In
the second part, you will do the same for semantic segmentation. via the Kaggle competition. In the third part, you will try to find and exploit the weaknesses of your classification and/or segmentation network. For the latter there is no competition format, but we do expect you to put significant effort in achieving
good performance on the self-posed goal for that part. Finally, we ask you to reflect and
produce an overall discussion with links to the lectures and “real world” computer vision.
In general, we will evaluate the correctness of your approach and your understanding of what you have done
that you demonstrate in the descriptions and discussions in the final notebook.
!!! : The codes are basically done, performing a really poor accuracy. The intended goal is to make optimization over them and specify the changes and the reasons.
The adversarial attack has errors. does not run.