Add a tool to extract features from OSM to retrain

src/train/coord_walker.py

@@ -0,0 +1,11 @@
+from src.detection.walker import Walker
+
+
+class CoordWalker(Walker):
+    def __init__(self, tile, nodes, square_image_length=50, zoom_level=19, step_width=0.66):
+        super(CoordWalker, self).__init__(tile, square_image_length, zoom_level, step_width)
+        self.nodes = nodes
+
+    def get_tiles(self):
+        squared_tiles = self._get_squared_tiles(self.nodes)
+        return squared_tiles

src/train/fetch.py

@@ -0,0 +1,52 @@
+from src.base.tile import Tile
+from src.base.bbox import Bbox
+from src.base.node import Node
+from src.base.tag import Tag
+
+from src.data.orthofoto.wms.wms_api import WmsApi
+
+from src.train.coord_walker import CoordWalker
+from src.train.osm_object_walker import OsmObjectWalker
+
+import argparse
+
+def main(args):
+    coords = list(map(lambda c: Node(*map(float, c.split(','))), args.coord))
+    bbox = Bbox.from_nodes(coords[0], coords[1])
+    if args.tags:
+        tags = map(lambda k, v: Tag(key=k, value=v), map(lambda kv: kv.split('=', 1), args.tags.split(',')))
+        #walker = OsmObjectWalker(Tile(image_api=WmsApi(), bbox=bbox), Tag(key='public_transport', value='platform'), square_image_length=100)
+        walker = OsmObjectWalker(Tile(image_api=WmsApi(), bbox=bbox), tags, square_image_length=100)
+    else:
+        walker = CoordWalker(Tile(image_api=WmsApi(), bbox=bbox), coords, square_image_length=100)
+
+    tiles = walker.get_tiles()
+    for n, t in enumerate(tiles):
+        centre_node = t.get_centre_node()
+        name = "fetch/{0:02.8}_{1:02.8}.png".format(centre_node.latitude, centre_node.longitude)
+        t.image.save(name, "PNG")
+        print(name)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        '--tags',
+        type=str,
+        default=None,
+        help='Tag to fetch from OSM: highway=crossing.'
+    )
+
+    parser.add_argument(
+        'coord',
+        type=str,
+        action='store',
+        nargs='+',
+        help='lon,lat coord in WGS84, if --tags bbox left,bottom right,top, else list of coords to fetch.')
+
+    args = parser.parse_args()
+    main(args)
+
+# mapproxy-util serve-develop mapproxy.yml
+# montage *.png -geometry 100x100+1+1 out.png
+# python retrain.py --image_dir retrain-data --print_misclassified_test_images

src/train/osm_object_walker.py

@@ -0,0 +1,25 @@
+from src.base.node import Node
+from src.detection.walker import Walker
+from src.data.osm.overpass_api import OverpassApi
+
+
+class OsmObjectWalker(Walker):
+    def __init__(self, tile, tags, square_image_length=50, zoom_level=19, step_width=0.66):
+        super(OsmObjectWalker, self).__init__(tile, square_image_length, zoom_level, step_width)
+        self.tags = tags
+
+    def get_tiles(self):
+        nodes = self._calculate_tile_centres()
+        squared_tiles = self._get_squared_tiles(nodes)
+        return squared_tiles
+
+    def _calculate_tile_centres(self):
+        centers = []
+
+        # [out:csv(::lat,::lon)][timeout:25];node["public_transport"="platform"]({{bbox}});out;
+        self.api = OverpassApi()
+        data = self.api.get(self.tile.bbox, self.tags, nodes=True, ways=False, relations=False, responseformat='csv(::lat,::lon)')
+        data = list(map(lambda cc: Node(float(cc[0]), float(cc[1])), data[1:]))
+        print(data)
+
+        return data