IDEA-Research / MP-Former

[CVPR 2023] Official implementation of the paper: MP-Former: Mask-Piloted Transformer for Image Segmentation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

MP-Former

This is the official implementation of the paper "MP-Former: Mask-Piloted Transformer for Image Segmentation". Accepted to CVPR 2023.

News

  • 8/19/2023: Code is released.

Install

The installation is exactly same as Mask2Former

Train

The following command is for no noise and all-layer MP training setting for instance segmentation.

bash run_50ep_no_noise_all_ly.sh

The following command is for no noise and all-layer MP training setting for panoptic segmentation.

bash run_50ep_no_noise_all_ly_panoptic.sh

Eval

You can set "weights" to the path of the checkpoint.

bash eval.sh

Model

This checkpoint is the 12-epoch checkpoint (last row in Table 7 of the paper). You are expected to have AP: 40.15 if you evaluate it.

About

[CVPR 2023] Official implementation of the paper: MP-Former: Mask-Piloted Transformer for Image Segmentation

License:Other


Languages

Language:Python 91.7%Language:Cuda 7.1%Language:C++ 0.8%Language:Shell 0.4%