Project-MONAI / Project-MONAI/MONAI
Create memory-friendly layers in MONAI to enable 3D large tensors
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 8.7k
- Forks
- 1.6k
- Avg merge
- 5d 1h
- Merged PRs (30d)
- 20
Description
Is your feature request related to a problem? Please describe.
Out-of-memory is a common issue. I will try to create new layers to enable memory-efficient forward pass (potentially backward as well).
Describe the solution you'd like
- Split the forward pass of feature maps in convolution into a for loop (PR #7215).
- Split the forward pass of feature maps in normalization into a for loop (PR #7215).
This idea has some similarities to the work below.
https://ieeexplore.ieee.org/abstract/document/9178453
Describe alternatives you've considered
- Split the forward pass of feature maps in convolution across multiple GPUs. (PR #7232)
- Split the forward pass of feature maps in normalization across multiple GPUs (PR #7232).
Additional context
N/A.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Review PR #7215 for the proposed convolution and normalization feature-map splitting, then compare PR #7232 for the multi-GPU alternatives. Determine whether the memory-friendly layers for large 3D tensors, including the intended forward pass and any backward-pass support, are already covered before proposing further work.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 20/100