/FCtL

[ICCV 2021] "From Contexts to Locality: Ultra-high Resolution Image Segmentation via Locality-aware Contextual Correlation" by Qi Li, Weixiang Yang, Wenxi Liu, Yuanlong Yu, Shengfeng He

Primary LanguagePythonMIT LicenseMIT

From Contexts to Locality: Ultra-high Resolution Image Segmentation via Locality-aware Contextual Correlation

From Contexts to Locality: Ultra-high Resolution Image Segmentation via Locality-aware Contextual Correlation
Qi Li, Weixiang Yang, Wenxi Liu, Yuanlong Yu, Shengfeng He
Accepted to ICCV 2021

Abstract

Ultra-high resolution image segmentation has raised increasing interests in recent years due to its realistic applications. In this paper, we innovate the widely used high-resolution image segmentation pipeline, in which an ultra-high resolution image is partitioned into regular patches for local segmentation and then the local results are merged into a high-resolution semantic mask. In particular, we introduce a novel locality-aware contextual correlation based segmentation model to process local patches, where the relevance between local patch and its various contexts are jointly and complementarily utilized to handle the semantic regions with large variations. Additionally, we present a contextual semantics refinement network that associates the local segmentation result with its contextual semantics, and thus is endowed with the ability of reducing boundary artifacts and refining mask contours during the generation of final high-resolution mask. Furthermore, in comprehensive experiments, we demonstrate that our model outperforms other state-of-the-art methods in public benchmarks.
tease

Method

framework

Test and train

Our code is based on GLNet
python>=3.6 and pytorch>=1.2.0
Please install the dependencies: pip install -r requirements.txt

dataset

Please register and download Inria Aerial dataset
Create folder named 'data_1', its structure is

data_1/
├── train
   ├── Sat
      ├── xxx_sat.tif
      ├── ...
   ├── Label
      ├── xxx_mask.png(two values:0-1)
      ├── ...
├── crossvali
├── offical_crossvali

test

Please download following pretrianed-model here
1.all.epoch.pth 2.B10.epoch.pth 3.B15.epoch.pth
bash test.sh

train

Please sequentially finish the following steps:
1.bash train_pre.sh(not necessary)
2.bash train_B10.sh(get medium context)
3.bash train_B15.sh(get large context)
4.bash train.sh

Results

DeepGlobe

result

Inria Aerial

result1

Citation

If you use this code or our results for your research, please cite our paper.

@inproceedings{li2021contexts,
  title={From Contexts to Locality: Ultra-high Resolution Image Segmentation via Locality-aware Contextual Correlation},
  author={Li, Qi and Yang, Weixiang and Liu, Wenxi and Yu, Yuanlong and He, Shengfeng},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={7252--7261},
  year={2021}
}