AgentTR

This is a PyTorch implementation of ECCV2022 paper Adaptive Agent Transformer for Few-Shot Segmentation

Usage

Environment

Python==3.8
torch==1.8.1
torchvision==0.9.0

Build Dependencies of Deformable DETR for Representation Encoder

cd model/ops/
bash make.sh
cd ../../

Data Preparation

• PASCAL-5^i

• COCO-20^i

• Please refer to CyCTR_Pytorch to prepare the PASCAL and COCO dataset for few-shot segmentation.

Train

Download the ImageNet pretrained backbones and put them into the initmodel directory.

Then, run this command:

    sh train.sh "dataset" "model_config"

Test

• Run the following command:

    sh test.sh "dataset" "model_config"

Acknowledgement

This project is built upon CyCTR_Pytorch and ASGNet, thanks for their great works!