In this course you will practice training deep neural network to solve severals computer vision problems, e.g. Image classification, Instance segmentation. You can choose the deep learning framework which you are familiar with. If you are a newbie in deep learning framework, we recommend you learn Pytorch. We highly encourage students who are not familiar with Python or Pytorch/Keras/TensorFlow to complete the following tutorials first.
The homeworks are due at 11:59pm. The due dates for each homeworks are on the syllabus page.
We will deduct a late penalty of 20% per additional late day.
Most of homeworks need you attend a private Kaggle competition, you will have to make your submission before deadline.
Also mail your reports to TA Jimmy before deadline at [email protected] with format CS_IOC5008_<STUDENT-ID>_HW<NUMBER>, e.g. CS_IOC5008_0656001_HW2.
Your HW should include
- Reports (in PDF)
- GitHub/GitLab repository link
- Introduction
- Methodology
- Findings or Summary
- Results (Please see the README in each HW folder for details)
- Model performance (70%): You can get at least 56% (70%x0.8) by scoring over the baseline (on Kaggle learderbaord) for each HW. Rank top 3 on leaderboard will be invited to make a presentation to share your methodology and get bonus on your final score
- Coding (10%): Python coding style should follow PEP8 for readability and your results should be reproducible.
- Reports (20%): Well-document the data summary, methodology or any findings in this HW
You may need GPU to accelerate the training of deep nenural network. We provide several free GPU resources for you, some of resources need registration and limited by usage.
- Google Colab: Free GPU usage for continous 24 hours
- FloydHub: Registration for free GPU trials
- Microsoft Azure): Registration for free GPU trials