Skip to content

serendipity999/deepglobe_land_cover_classification_with_deeplabv3

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DeepGlobe Land Cover Classification Challenge 土地利用分类竞赛

Timeline 工作时间线

  • Dataset 数据集准备
    • Load file path文件目录导入
    • 底图与标注文件的对应关系
    • 标注文件的格式化
    • one-hot 编码
  • DeepLab 模型准备
    • 迁移学习
  • Test 测试
  • Result 结果展示

DATASET 数据集

DATA 数据

  • The training data for Land Cover Challenge contains 803 satellite imagery in RGB, size 2448x2448.
  • The imagery has 50cm pixel resolution, collected by DigitalGlobe's satellite.
  • You can download the training data in the download page with filetype of “Starting Kit”. Testing satellite images will be will be uploaded later.
  • 训练数据集包括803张卫星图片,RGB格式,尺寸2448 * 2448
  • 图像分辨率为50cm,由 DigitalGlobe's 卫星提供
  • 可通过下载页点击"Starting Kit"下载数据。

Label 标注

  • Each satellite image is paired with a mask image for land cover annotation. The mask is a RGB image with 7 classes of labels, using color-coding (R, G, B) as follows.

    • Urban land: 0,255,255 - Man-made, built up areas with human artifacts (can ignore roads for now which is hard to label)
    • Agriculture land: 255,255,0 - Farms, any planned (i.e. regular) plantation, cropland, orchards, vineyards, nurseries, and ornamental horticultural areas; confined feeding operations.
    • Rangeland: 255,0,255 - Any non-forest, non-farm, green land, grass
    • Forest land: 0,255,0 - Any land with x% tree crown density plus clearcuts.
    • Water: 0,0,255 - Rivers, oceans, lakes, wetland, ponds.
    • Barren land: 255,255,255 - Mountain, land, rock, dessert, beach, no vegetation
    • Unknown: 0,0,0 - Clouds and others
  • File names for satellite images and the corresponding mask image are _sat.jpg and _mask.png. is a randomized integer.

  • Please note:

    • The values of the mask image may not be the exact target color values due to compression. When converting to labels, please binarize each R/G/B channel at threshold 128.
    • Land cover segmentation from high-resolution satellite imagery is still an exploratory task, and the labels are far from perfect due to the cost for annotating multi-class segmentation mask. In addition, we intentionally didn't annotate roads because it's already covered in a separate road challenge.
  • 每张卫星图片有一张与之对应的标注图片。这张标注图片也是RGB格式,一共分为7类,每类对应的图像(R,G,B)编码对应关系如下:

    • 城市土地: 0,255,255,浅蓝色,人造建筑(可以忽略道路)
    • 农业用地: 255,255,0,黄色,农田,任何计划中(定期)的种植、农田、果园、葡萄园、苗圃、观赏性园艺以及养殖区
    • 牧场: 255,0,255,紫色,除了森林,农田之外的绿色土地,草地
    • 森林:0,255,0,绿色,任何土地上有x%的树冠密度。
    • 水系:0,0,255,深蓝色,江河湖海湿地
    • 荒地:255,255,255, 白色,山地,沙漠,戈壁,沙滩,没有植被的地方
    • 未知土地: 0,0,0,黑色,云层遮盖或其他因素
  • 卫星图片和与其对应的标注图片的命名格式为<id>_sat.jpg和<id>_mask.png,<id>是一个随机的整数。

  • 需要注意:

    • 由于压缩,标注图像的值可能不是准确的目标颜色值。当转换到标签时,请将每个R/G/B通道按128阈值二值化。
    • 高分辨率卫星图像的土地覆被分割仍然是一个探索性的任务,由于标注多类分割的代价很大,标签还远远不够完善。此外,我们故意不标注道路,因为它已经包含在一个单独的道路提取挑战赛中。

Evaluation Metric 评价指标

  • We will use pixel-wise mean Intersection over Union (mIoU) score as our evaluation metric.
    • IoU is defined as: True Positive / (True Positive + False Positive + False Negative).
    • mean IoU is calculated by averaging over all classes.
  • Please note the Unknown class (0,0,0) is not an active class used in evaluation. Pixels marked as Unknown will simply be ignored. So effectively mIoU is averaging over 6 classes in total.
  • 我们将采用像素级别的平均IoU分数作为评价指标。
    • IoU的定义是:交集/并集,公式: 预测准确的面积/(预测准确面积 + 没有预测出来的面积 + 预测错误的面积)
    • 平均IoU由各个类别的IoU取均值得到
  • 需要注意的是"未知土地"(0,0,0)并不是一个真实的类别,在这个计算中也没有起到作用。所以有效的mIoU是前6个类别IoU的均值。

Result 结果展示

About

DeepGlobe Land Cover Classification Challenge

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%