pytorch / pytorch/vision

Mask R-CNN with MobileNet v3 backbone

Open
#4,048 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

🚀 Feature

Add Mask R-CNN with MobileNet v3 backbone

Motivation

In the recent release torchvision added support for mobilenet v3 and extended it to work with instance segmentation models.
torchvision.models.detection.fasterrcnn_mobilenet_v3_large_fpn
However, support for Mask R-CNN with MobileNet v3 backbone is still missing and would be a worthy addition.

Pitch

  • Similar module like torchvision.models.detection.fasterrcnn_mobilenet_v3_large_fpn but to work with Mask R-CNN like resnet has torchvision.models.detection.maskrcnn_resnet50_fpn
  • MobileNet V3 would make inference much faster
  • Mask R-CNN is widely used owing to its dual use of getting mask+bounding box

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 comparing the existing torchvision.models.detection.fasterrcnn_mobilenet_v3_large_fpn and torchvision.models.detection.maskrcnn_resnet50_fpn entry points and their surrounding implementation. Trace how the MobileNet V3 backbone is connected to the instance-segmentation components. Done means a Mask R-CNN entry point using MobileNet V3 supports both masks and bounding boxes with coverage comparable to the existing model.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-vision, machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.