StevenCyb / Sat_Segment_Dataset_Generator

Create a dataset for semantic segmentation of satellite images.

Sat-Segment-Dataset-Generator

This project was developed to create datasets for semantic segmentation with satellite images. The satellite images are obtained from Here-Map and the OSM data from OpenStreetMap. Depending on the configuration a mask is created, as for example in the following figure. When using the data, please take note of the license terms of the data sources. The image shows the satellite image on the left and the mask on the right. In this figure two variants of the mask are shown. In the left mask the OSM data was drawn normally, while on the right flood-fill was used to fill the street segments. Depending on the quality of the image data, the flood-fill can deliver good or bad results.

How To Use It

This section describes how you can use the tool. This example uses the scenario shown above.

Create The Configuration

The configuration defines how the dataset is supposed to be generated. The structure of the configuration is explained by the following comments.

{
/*
* This part defines the area for which a data set is to be defined.
* The range is indicated by the left upper and right lower corners.
* The fixed points are not necessarily strict. The following figure shows this using an example. 
* The green dots represent the defined corner points and the black rectangle the tiles.
* This means that instead of the defined orange rectangle, the orange highlighted area is included (under certain zoom-levels).
*
* The minimum and maximum zoom level can also be restricted. 
* If this is left at 0 and 19, all possible sizes for the range are automatically created.
*/
    "location_range": {
        "top_left_location": {
            "latitude": 48.779346,
            "longitude": 9.172896
        },
        "bottom_right_location": {
            "latitude": 48.774943,
            "longitude": 9.184387
        },
        "zoom_range": {
            "min": 0,
            "max": 19
        }
    },
/*
* Defines the image size of the Tiles. You can choose between 256 or 512.
*/
	"here_api": {
		"tile_size": 512
	},
/*
* This part determines timeouts and attempts for the Overpass-Api.
*/
    "overpass_api": {
		"max_tries": 5,
		"connection_establish_timeout": 5,
		"response_timeout": 30 
	},
/*
* Which categories are needed and how they are drawn is defined here.
* For each category a query, the extension radius and the drawing options are specified.
* The current range of the query is defined with "{{bbox}}", so this should be included in every query.
* The expansion factor prevents objects at the edge from being truncated by the query.
* The color of the category and type which can be "area", "line" or "dot" is always specified in the drawing options.
* If a category consists of areas and lines, it must be defined twice with the same color as in the example water and highway.
* Depending on the type, there are additional options.
* - Area: There are no additional options here.
* - Line: The line thickness can also be specified here.
* - Dot: The radius can also be defined here.
* 
* Additionally flood-fill can be added to all types. The specified value determines the threshold.
* An example can be seen in the second category for highway.
*/
	"categories": [
	    {
			"overpass_api_query": "way[area][highway]({{bbox}}); way[area][amenity=parking]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "area",
				"color": {"b":0, "g":255, "r":0}
			}
	    },
	    {
			"overpass_api_query": "way[!area][highway]({{bbox}}); way[!area][amenity=parking]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "line",
				"color": {"b":0, "g":255, "r":0},
				"line_width": 3,
				"flood_fill": 20
			}
	    },
	    {
			"overpass_api_query": "way[area][natural=water]({{bbox}}); way[area][natural=dam]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "area",
				"color": {"b":255, "g":0, "r":0}
			}
	    },
	    {
			"overpass_api_query": "way[waterway=river]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "line",
				"color": {"b":255, "g":0, "r":0},
				"line_width": 3
			}
	    },
	    {
			"overpass_api_query": "way[railway]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "line",
				"color": {"b":0, "g":0, "r":255},
				"line_width": 3
			}
	    },
	    {
			"overpass_api_query": "node[natural=tree]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "dot",
				"color": {"b":0, "g":0, "r":0},
				"circuit_radius": 3
			}
	    },
	    {
			"overpass_api_query": "way[building]({{bbox}});",
			"expand_view": 0.0005,
			"draw_options": {
				"type": "area",
				"color": {"b":255, "g":255, "r":255}
			}
	    }
	]
}

Run The SSDG

After the configuration has been created, the tool can be started with the following command, where --output_path defines the location to store the dataset and --config_path the path to the configuration created above. You must replace {{YOUR_APP_ID}} and {{YOUR_APP_CODE}} with your Here-Api credentials, that you can get here.

python3 sat_segment_dataset_generator.py --here_app_id {{YOUR_APP_ID}} --here_app_code {{YOUR_APP_CODE}} --output_path ./dataset --config_path ./config.cfg