pip install -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu116
- run on HPC
jupyter notebook --no-browser --port=40000 --ip=$HOSTNAME - run on local computer
ssh -i path/to/sshkey USERNAME@l1.hpc.dtu.dk -g -L8080:HOSTNAME:PORT -N
- Thursday 5th. at 15
- Monday 9th. at 14
- Thursday 12th. at 14
-
Phase 1
- Get familiar with Monai by classifing MedNIST data using deep learning. Comparison between two or more deep learning methods for 3D classification of the six 3D datasets.
- Investigate tools network visualisation e.g. GradCAM
- Apply the network visualisation to the MedNIST models
-
Phase 2
- Structure the insect data for deep learning-based 3D classification (sort, scale, etc.).
- Implement the same deep learning-based classification models for the insect dataset as used for the MedNIST
- Implement one or more improved deep learning models for the insect dataset
- Verify the data by applying network visualisation to insect model
- Create requirement file (so all our envs are the same)
- Make sure all have acess to HPC and know how to use it
- Create a problem description
- Is hand-in only a slide deck with comments, and how elaborate should the comments be?
- Since there is only 500 images, should we do some image augmentation, and which methods would be a good idea, rotating, scaling, bluring?
- Is it relevant to train models for all 6 3D MedNIST datasets?
- Is it nessesary to use the monai models?