{
  "id": 26655,
  "title": "GeoJSON has more data than WKT",
  "url": "/competitions/dstl-satellite-imagery-feature-detection/discussion/26655",
  "author_name": "",
  "post_date": "2016-12-18T21:02:32.067Z",
  "votes": 3,
  "comment_count": 1,
  "views": 247,
  "content": "<p>There is more amount of, and more specific, data inside some of the geoJSON files than the WKT csv. For example in image 6040_1_0 the geoJSON has 2 path classes, FOOTPATH_TRAIL and POOR_DIRT_CART_TRACK, while the WKT has the same features mapped as just class 4. (Also, in the goeJSON there is nothing in the metadata that associates these with class 4). </p>\n\n<p>In 6040_1_3, the geoJSON has layers called SCRUBLAND and DEMARCATED_NON_CROP which aren't in the WKT. With these additions all the area of 6040_1_3 is accounted for in the geoJSON. In most of the other images there are gaps with no label. Like in 6040_1_0 I can see its mostly scrubland, but it doesn't have that label in either the geoJSON or WKT. But from looking at 6040_1_3 it looks like that data is available? Were we meant not to have a lot of the other labels that identify the entire image, or were they left out by mistake?</p>",
  "messages": [
    {
      "id": "151091",
      "postDate": "12/18/2016 21:02:32",
      "content": "<p>There is more amount of, and more specific, data inside some of the geoJSON files than the WKT csv. For example in image 6040_1_0 the geoJSON has 2 path classes, FOOTPATH_TRAIL and POOR_DIRT_CART_TRACK, while the WKT has the same features mapped as just class 4. (Also, in the goeJSON there is nothing in the metadata that associates these with class 4). </p>\n\n<p>In 6040_1_3, the geoJSON has layers called SCRUBLAND and DEMARCATED_NON_CROP which aren't in the WKT. With these additions all the area of 6040_1_3 is accounted for in the geoJSON. In most of the other images there are gaps with no label. Like in 6040_1_0 I can see its mostly scrubland, but it doesn't have that label in either the geoJSON or WKT. But from looking at 6040_1_3 it looks like that data is available? Were we meant not to have a lot of the other labels that identify the entire image, or were they left out by mistake?</p>",
      "rawMarkdown": "There is more amount of, and more specific, data inside some of the geoJSON files than the WKT csv. For example in image 6040_1_0 the geoJSON has 2 path classes, FOOTPATH_TRAIL and POOR_DIRT_CART_TRACK, while the WKT has the same features mapped as just class 4. (Also, in the goeJSON there is nothing in the metadata that associates these with class 4). \r\n\r\nIn 6040_1_3, the geoJSON has layers called SCRUBLAND and DEMARCATED_NON_CROP which aren't in the WKT. With these additions all the area of 6040_1_3 is accounted for in the geoJSON. In most of the other images there are gaps with no label. Like in 6040_1_0 I can see its mostly scrubland, but it doesn't have that label in either the geoJSON or WKT. But from looking at 6040_1_3 it looks like that data is available? Were we meant not to have a lot of the other labels that identify the entire image, or were they left out by mistake?",
      "votes": null
    },
    {
      "id": "151124",
      "postDate": "12/19/2016 00:00:40",
      "content": "<p>Regarding the first question: on the DATA page in the bottom you may find a dictionary with object to class correspondence. Here is a copy:</p>\n\n<p>filename_to_classType = {</p>\n\n<p>'001_MM_L2_LARGE_BUILDING':1,</p>\n\n<p>'001_MM_L3_RESIDENTIAL_BUILDING':1,</p>\n\n<p>'001_MM_L3_NON_RESIDENTIAL_BUILDING':1,</p>\n\n<p>'001_MM_L5_MISC_SMALL_STRUCTURE':2,</p>\n\n<p>'002_TR_L3_GOOD_ROADS':3,</p>\n\n<p>'002_TR_L4_POOR_DIRT_CART_TRACK':4,</p>\n\n<p>'002_TR_L6_FOOTPATH_TRAIL':4,</p>\n\n<p>'006_VEG_L2_WOODLAND':5,</p>\n\n<p>'006_VEG_L3_HEDGEROWS':5,</p>\n\n<p>'006_VEG_L5_GROUP_TREES':5,</p>\n\n<p>'006_VEG_L5_STANDALONE_TREES':5,</p>\n\n<p>'007_AGR_L2_CONTOUR_PLOUGHING_CROPLAND':6,</p>\n\n<p>'007_AGR_L6_ROW_CROP':6, </p>\n\n<p>'008_WTR_L3_WATERWAY':7,</p>\n\n<p>'008_WTR_L2_STANDING_WATER':8,</p>\n\n<p>'003_VH_L4_LARGE_VEHICLE':9,</p>\n\n<p>'003_VH_L5_SMALL_VEHICLE':10,</p>\n\n<p>'003_VH_L6_MOTORBIKE':10\n}</p>",
      "rawMarkdown": "Regarding the first question: on the DATA page in the bottom you may find a dictionary with object to class correspondence. Here is a copy:\r\n\r\nfilename_to_classType = {\r\n\r\n'001_MM_L2_LARGE_BUILDING':1,\r\n\r\n'001_MM_L3_RESIDENTIAL_BUILDING':1,\r\n\r\n'001_MM_L3_NON_RESIDENTIAL_BUILDING':1,\r\n\r\n'001_MM_L5_MISC_SMALL_STRUCTURE':2,\r\n\r\n'002_TR_L3_GOOD_ROADS':3,\r\n\r\n'002_TR_L4_POOR_DIRT_CART_TRACK':4,\r\n\r\n'002_TR_L6_FOOTPATH_TRAIL':4,\r\n\r\n'006_VEG_L2_WOODLAND':5,\r\n\r\n'006_VEG_L3_HEDGEROWS':5,\r\n\r\n'006_VEG_L5_GROUP_TREES':5,\r\n\r\n'006_VEG_L5_STANDALONE_TREES':5,\r\n\r\n'007_AGR_L2_CONTOUR_PLOUGHING_CROPLAND':6,\r\n\r\n'007_AGR_L6_ROW_CROP':6, \r\n\r\n'008_WTR_L3_WATERWAY':7,\r\n\r\n'008_WTR_L2_STANDING_WATER':8,\r\n\r\n'003_VH_L4_LARGE_VEHICLE':9,\r\n\r\n'003_VH_L5_SMALL_VEHICLE':10,\r\n\r\n'003_VH_L6_MOTORBIKE':10\r\n}",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 151124,
      "author_name": "alexlu",
      "author_url": "",
      "post_date": "12/19/2016 00:00:40",
      "content": "<p>Regarding the first question: on the DATA page in the bottom you may find a dictionary with object to class correspondence. Here is a copy:</p>\n\n<p>filename_to_classType = {</p>\n\n<p>'001_MM_L2_LARGE_BUILDING':1,</p>\n\n<p>'001_MM_L3_RESIDENTIAL_BUILDING':1,</p>\n\n<p>'001_MM_L3_NON_RESIDENTIAL_BUILDING':1,</p>\n\n<p>'001_MM_L5_MISC_SMALL_STRUCTURE':2,</p>\n\n<p>'002_TR_L3_GOOD_ROADS':3,</p>\n\n<p>'002_TR_L4_POOR_DIRT_CART_TRACK':4,</p>\n\n<p>'002_TR_L6_FOOTPATH_TRAIL':4,</p>\n\n<p>'006_VEG_L2_WOODLAND':5,</p>\n\n<p>'006_VEG_L3_HEDGEROWS':5,</p>\n\n<p>'006_VEG_L5_GROUP_TREES':5,</p>\n\n<p>'006_VEG_L5_STANDALONE_TREES':5,</p>\n\n<p>'007_AGR_L2_CONTOUR_PLOUGHING_CROPLAND':6,</p>\n\n<p>'007_AGR_L6_ROW_CROP':6, </p>\n\n<p>'008_WTR_L3_WATERWAY':7,</p>\n\n<p>'008_WTR_L2_STANDING_WATER':8,</p>\n\n<p>'003_VH_L4_LARGE_VEHICLE':9,</p>\n\n<p>'003_VH_L5_SMALL_VEHICLE':10,</p>\n\n<p>'003_VH_L6_MOTORBIKE':10\n}</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "151091": "There is more amount of, and more specific, data inside some of the geoJSON files than the WKT csv. For example in image 6040_1_0 the geoJSON has 2 path classes, FOOTPATH_TRAIL and POOR_DIRT_CART_TRACK, while the WKT has the same features mapped as just class 4. (Also, in the goeJSON there is nothing in the metadata that associates these with class 4). \r\n\r\nIn 6040_1_3, the geoJSON has layers called SCRUBLAND and DEMARCATED_NON_CROP which aren't in the WKT. With these additions all the area of 6040_1_3 is accounted for in the geoJSON. In most of the other images there are gaps with no label. Like in 6040_1_0 I can see its mostly scrubland, but it doesn't have that label in either the geoJSON or WKT. But from looking at 6040_1_3 it looks like that data is available? Were we meant not to have a lot of the other labels that identify the entire image, or were they left out by mistake?",
    "151124": "Regarding the first question: on the DATA page in the bottom you may find a dictionary with object to class correspondence. Here is a copy:\r\n\r\nfilename_to_classType = {\r\n\r\n'001_MM_L2_LARGE_BUILDING':1,\r\n\r\n'001_MM_L3_RESIDENTIAL_BUILDING':1,\r\n\r\n'001_MM_L3_NON_RESIDENTIAL_BUILDING':1,\r\n\r\n'001_MM_L5_MISC_SMALL_STRUCTURE':2,\r\n\r\n'002_TR_L3_GOOD_ROADS':3,\r\n\r\n'002_TR_L4_POOR_DIRT_CART_TRACK':4,\r\n\r\n'002_TR_L6_FOOTPATH_TRAIL':4,\r\n\r\n'006_VEG_L2_WOODLAND':5,\r\n\r\n'006_VEG_L3_HEDGEROWS':5,\r\n\r\n'006_VEG_L5_GROUP_TREES':5,\r\n\r\n'006_VEG_L5_STANDALONE_TREES':5,\r\n\r\n'007_AGR_L2_CONTOUR_PLOUGHING_CROPLAND':6,\r\n\r\n'007_AGR_L6_ROW_CROP':6, \r\n\r\n'008_WTR_L3_WATERWAY':7,\r\n\r\n'008_WTR_L2_STANDING_WATER':8,\r\n\r\n'003_VH_L4_LARGE_VEHICLE':9,\r\n\r\n'003_VH_L5_SMALL_VEHICLE':10,\r\n\r\n'003_VH_L6_MOTORBIKE':10\r\n}"
  },
  "source": "meta"
}