pytorch / pytorch/vision

I hope you can provide 3D align-pooling layer directly. Pray.

Open
#2,774 4 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

module: ops new feature topic: object detection
Dominant language
Python
Stars
17.9k
Forks
7.3k
Avg merge
1d 15h
Merged PRs (30d)
13

Description

I see that 2D align-pooling has been implemented, but there is no 3D one. I do target detection of 3D voxels. The input data format is (batch, channel, depth, height, w). I customized a 3D align-pooling Op. However, there is a big problem in the deployment. The custom OP needs to be registered in the torchscript. In this way, the model can be exported normally, and finally the model can be loaded on the C + + side.I tried for months, but it didn't work. I hope you can provide 3D align-pooling layer directly. Pray.

cc @fmassa @vfdev-5

Contributor guide

Open the contributing guide

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

Start by reviewing the existing 2D align-pooling implementation and the TorchScript custom-operator registration path. Check the requested (batch, channel, depth, height, width) voxel input and C++ loading workflow; done means a 3D align-pooling layer can be exported and loaded without a custom deployment workaround.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, python
Domain
computer-vision, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.