onnx / onnx/models

Mask-RCNN network contains concat with one input

Open
#214 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Jupyter Notebook
Stars
9.8k
Forks
1.6k
PR merge metrics
No merged PRs in 30d

Description

After examining the Mask-RCNN onnx file obtained from the ONNX Model zoo:
https://github.com/onnx/models/tree/master/vision/object_detection_segmentation/mask-rcnn

I noticed a few structures that look like this:
Unsqueeze -> Concat -> ConstantOfShape
An example can be nodes '685' -> '686' -> '687'.

I see that there is a 'Concat' operation that only obtains one tensor. It would seem to be redundant. Is this necessary to the model? If not, can it be removed from the model or would this be a useful pass to implement?

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Inspect the Mask-RCNN ONNX file linked in the issue, focusing on nodes 685–687 and the Unsqueeze → Concat → ConstantOfShape chain. Check the ONNX graph and operator rules to determine whether the single-input Concat is necessary; done means documenting the finding and, if appropriate, identifying a validated removal pass.

Written by the indexing model from the issue text.

Assessment

Domain
computer-vision, machine-learning
Issue type
Refactor
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.