Data

Tools

Indexes

Applications

Experiments

Open Source Science

farmland-instance-segmentation

instance segmentation in IFLYTEK farmland challenges based on mmdetection

https://github.com/wangzehui20/farmland-instance-segmentation

Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ○
    Academic publication links
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (10.5%) to scientific vocabulary

Keywords

farmland instance-segmentation mask-rcnn mmdetection remote-sensing
Last synced: 6 months ago · JSON representation ·

Repository

instance segmentation in IFLYTEK farmland challenges based on mmdetection

Basic Info
  • Host: GitHub
  • Owner: wangzehui20
  • License: apache-2.0
  • Language: Python
  • Default Branch: master
  • Homepage:
  • Size: 11.5 MB
Statistics
  • Stars: 2
  • Watchers: 1
  • Forks: 3
  • Open Issues: 1
  • Releases: 0
Topics
farmland instance-segmentation mask-rcnn mmdetection remote-sensing
Created over 4 years ago · Last pushed over 4 years ago
Metadata Files
Readme Contributing License Code of conduct Citation

README.md

Farmland Segmentation on IFLYTEK challenge

In this Farmland Segmentation Challenge, I only use Mask RCNN + FPN on MMDetection to achieve rank 3

My code is mainly located in experiment_farmland folder

Introduction

This competition aims to extract farmland segmentation from large remote-sensing images, so i will introduce my plan explicitly later

MMDetection is the most convenient and useful open source framework to learn deep learning, you could achieve better scores easily from most projects and study its poetical code. Appreciating to the contributors of MMDetection!

### Installation

Firstly, you need to configure MMDetection environment

  • I recommend you to follow official guide
  • and also refer to requirements from experiment_farmland/requirements.txt
  • conda install gdal

Custom dataset

Competition official gives large remote-sensing images that needs to be clipped into the small images and MMDetection requires COCO dataset format and so on

I implement above content by sliding window overlapping clipping in experiment_farmland\mask_rcnn\1024_256\dataset, image size is 1024 and overlap is 256 pixels

Your dataset transformed to coco dataset after that soft linked experiment_farmland\mask_rcnn\data_1024_256

Then, replace classes with your classes before training:

  • mmdet/datasets/coco.py

CoCoDataset(CustomDataset): CLASSES = ('farm_land')

  • mmdet/core/evaluation/class_names.py

def coco_classes(): return ['farm_land']

  • experiment_farmland\mask_rcnn\1024_256\code\mask_rcnn_res50\coco_instance.py

classes=('farm_land')

Train and Test

experiment_farmland\mask_rcnn

```

Train

bash dist_train.sh

Test

bash dist_test.sh ```

Tricks

  • GIoU Loss
  • Soft NMS

Post process

Q: Segmentation exist overlapping which may generate from inferior inference results and resume origin remote-sensing image from small clipped images.

A: My solution is that union IoU > 0.5 or intersection/polygon(low score) > 0.7 which help me improve score approximately 5 points in semi-finals

Demo

`experiment_farmland\mask_rcnn\1024_256\demo\demo.py

It could generate gt, pred, pred after post process images to analysis problem

  • pred and gt, better in regular farmland

avatar

  • pred and predpost

avatar

Also, above demo contrast tools refers to my another work, BatchLabelCrop

:smile::stuckouttonguewinkingeye: :kissingsmilingeyes:

Owner

  • Login: wangzehui20
  • Kind: user
  • Company: University of Chinese Academy of Sciences

Citation (CITATION.cff) 

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
authors:
  - name: "MMDetection Contributors"
title: "OpenMMLab Detection Toolbox and Benchmark"
date-released: 2018-08-22
url: "https://github.com/open-mmlab/mmdetection"
license: Apache-2.0

GitHub Events

Total
  • Watch event: 1
Last Year
  • Watch event: 1

Dependencies

.github/workflows/build.yml actions
  • actions/checkout v2 composite
  • actions/setup-python v2 composite
  • codecov/codecov-action v1.0.10 composite
.github/workflows/build_pat.yml actions
  • actions/checkout v2 composite
.github/workflows/deploy.yml actions
  • actions/checkout v2 composite
  • actions/setup-python v2 composite
docker/Dockerfile docker
  • pytorch/pytorch ${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel build
docker/serve/Dockerfile docker
  • pytorch/pytorch ${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel build
experiment_farmland/requirements.txt pypi
  • Shapely ==1.7.1
  • imgaug ==0.4.0
  • matplotlib ==3.4.0
  • mmcv ==1.3.14
  • mmcv_full ==1.3.8
  • mmdet ==2.16.0
  • numpy ==1.19.2
  • opencv_python ==4.5.1.48
  • osgeo ==0.0.1
  • pycocotools ==2.0
  • pygdal ==2.2.3.5
  • pyshp ==2.1.3
  • scikit_image ==0.18.1
  • skimage ==0.0
  • tabulate ==0.8.9
  • torch ==1.7.0
  • tqdm ==4.46.0
requirements/build.txt pypi
  • cython *
  • numpy *
requirements/docs.txt pypi
  • recommonmark *
  • sphinx *
  • sphinx_markdown_tables *
  • sphinx_rtd_theme *
requirements/mminstall.txt pypi
  • mmcv-full >=1.3.8
requirements/optional.txt pypi
  • albumentations >=0.3.2
  • cityscapesscripts *
  • imagecorruptions *
  • scipy *
  • sklearn *
requirements/readthedocs.txt pypi
  • mmcv *
  • torch *
  • torchvision *
requirements/runtime.txt pypi
  • matplotlib *
  • numpy *
  • pycocotools *
  • pycocotools-windows *
  • six *
  • terminaltables *
requirements/tests.txt pypi
  • asynctest * test
  • codecov * test
  • flake8 * test
  • interrogate * test
  • isort ==4.3.21 test
  • kwarray * test
  • onnx ==1.7.0 test
  • onnxruntime ==1.5.1 test
  • pytest * test
  • ubelt * test
  • xdoctest >=0.10.0 test
  • yapf * test
requirements.txt pypi
setup.py pypi