# BorderMask

**Repository Path**: shi-junyong/BorderMask

## Basic Information

- **Project Name**: BorderMask
- **Description**: This is the code and dataset used in the paper “BorderMask: Enhanced Boundary Perception and Streamlined Instance Segmentation”.
- **Primary Language**: Unknown
- **License**: MulanPSL-1.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2024-08-25
- **Last Updated**: 2024-08-29

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

###  # BorderMask


 **#### Repository Introduction.** 
The code and datasets used in the paper "BorderMask: Enhanced Boundary Perception and Streamlined Instance Segmentation" are provided here. Detailed code will be made available at this location once the paper is officially accepted for publication.

 **#### Algorithm Comparison and Project Environment.** 
We improve the Mask R-CNN and propose our BorderMask model. Unless otherwise specified, we use ResNet-50 and FPN as the backbone network. All comparative experiments are conducted using mixed precision on MMDetection, with the training environment set up on an Ubuntu 20.04 system, CUDA v11.8, cuDNN v8.6.0, and Pytorch 2.2.1, utilizing a single NVIDIA 3080Ti GPU. The learning rate (lr) schedule follows a multiplier of 0.008, and the optimizer employed is Stochastic Gradient Descent (SGD).

 **#### Installation Guide.** 
Refer to the official environment setup for Ubuntu 20.04 and PyTorch 2.2.1.

 **#### Datasets Used in the Paper** 
To better understand the significance of the three proposed components, we conduct ablation experiments on the widely used MS COCO 2017 dataset to evaluate BorderMask and compare it with other mainstream instance segmentation algorithms, demonstrating the effectiveness of BorderMask in boundary segmentation. Additionally, extensive comparative experiments are carried out on the PASCAL VOC 2012 and Cityscapes datasets to further validate the superiority of BorderMask.
MS COCO 2017 dataset download URL：https://cocodataset.org/#download
PASCAL VOC 2012 dataset download URL：http://host.robots.ox.ac.uk/pascal/VOC/voc2012/#data
Cityscapes dataset download URL：https://www.cityscapes-dataset.com/

 **#### Experimental evaluation.** 
In evaluating instance segmentation results, we use standard COCO-format evaluation metrics, including AP (IoU=0.5:0.95), AP50 (IoU=0.5), AP75 (IoU=0.75), APS (AP for small instances), APM (AP for medium instances), and APL (AP for large instances).

 **#### Code Description** 
The detailed code will be updated after the paper is published.