Segformer Semantic Segmentation #131
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Closed
danielrhuynh
requested changes
Jun 14, 2024
src/perception/semantic_segmentation/semantic_segmentation/segmentation_node.py
Show resolved
Hide resolved
src/perception/semantic_segmentation/semantic_segmentation/segmentation_node.py
Outdated
Show resolved
Hide resolved
src/perception/semantic_segmentation/semantic_segmentation/segmentation_node.py
Outdated
Show resolved
Hide resolved
src/perception/semantic_segmentation/resource/cityscapes_demo.png
Outdated
Show resolved
Hide resolved
src/perception/semantic_segmentation/launch/semantic_segmentation.launch.py
Outdated
Show resolved
Hide resolved
docker/perception/semantic_segmentation/semantic_segmentation.Dockerfile
Outdated
Show resolved
Hide resolved
docker/perception/semantic_segmentation/semantic_segmentation.Dockerfile
Outdated
Show resolved
Hide resolved
…:WATonomous/wato_monorepo into lereljic/segformer-semantic-segmentation
|
Wait for Eddie before merging tho |
|
Since this is a transformer, how much vram does this take up? |
Edwardius
requested changes
Jul 1, 2024
There was a problem hiding this comment.
Great work! A large portion of my concerns have to do with the size of the docker image, as well as the ros messages being used.
To reduce the size of the docker image, there are two avenues:
- wrap your head around docker caching to make MMLab stuff work
- use a simpler repo that has an implementation of segformer, ie. https://github.com/lucidrains/segformer-pytorch
As for ros messages:
- I would instead store the masks of each class for downstream processing, instead of a rgb image, this should be more efficient (by at least a couple orders of magnitude because you are passing arrays of boolean values vs RGB image which has 2^8 bits of data multiplied by the image size)
docker/perception/semantic_segmentation/semantic_segmentation.Dockerfile
Outdated
Show resolved
Hide resolved
docker/perception/semantic_segmentation/semantic_segmentation.Dockerfile
Outdated
Show resolved
Hide resolved
docker/perception/semantic_segmentation/semantic_segmentation.Dockerfile
Show resolved
Hide resolved
docker/perception/semantic_segmentation/semantic_segmentation.Dockerfile
Show resolved
Hide resolved
src/perception/semantic_segmentation/semantic_segmentation/segmentation_node.py
Show resolved
Hide resolved
src/perception/semantic_segmentation/semantic_segmentation/segmentation_node.py
Outdated
Show resolved
Hide resolved
src/perception/semantic_segmentation/semantic_segmentation/segmentation_node.py
Outdated
Show resolved
Hide resolved
|
I moved all the config stuff in the yaml file, the docker file should be smaller now. The publisher message is still one Image but it's just the mask now. I can change to a boolean array (such as ByteMultiArray I think) or the new vision message you mentioned. Let me know which one I should implement. |
Edwardius
approved these changes
Jul 30, 2024
|
Should a LabelInfo topic still be created for semantics, to map the class name to the int value in the mask? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Node running:
The model is: "segformer_mit-b2_8xb1-160k_cityscapes-1024x1024"
Segformer_node.mov