Postprocessing

Smoothing, combining etc.

Non-maximum suppression

First the commonly used NMS with bounding boxes, that prioritizes either confidence score (default) or bounding box area.

---

source

non_max_suppression_fast

def non_max_suppression_fast(
    boxes, scores, overlap_thresh:float, sort_criterion:str='score'
):

Possibility to sort boxes by score (default) or area

Non-max suppression can in theory be applied also on polygons, but it hasn’t been used in any publications as far as I know.

If non_max_suppression_poly is used to eliminate polygons, threshold might need to be smaller than typical value of 0.7 that is used.

---

source

poly_IoU

def poly_IoU(
    poly_1:Polygon, poly_2:Polygon
)->float:

Calculate IoU for two shapely Polygons

---

source

non_max_suppression_poly

def non_max_suppression_poly(
    geoms, scores, overlap_thresh:float, sort_criterion:str='score'
):

Do non-max suppression for shapely Polygons in geoms. Can be sorted according to area or score

Some utils to run above functions to GeoDataFrames

---

source

do_nms

def do_nms(
    gdf:GeoDataFrame, nms_thresh:float=0.7, crit:str='score'
)->GeoDataFrame:

Perform non-max suppression for bounding boxes using nms_threshold to gdf

def do_poly_nms(gdf:gpd.GeoDataFrame, nms_thresh=0.1, crit='score') -> gpd.GeoDataFrame:
    "Perform non-max suppression for polygons using `nms_threshold` to `gdf`"
    gdf = gdf.copy()
    scores = gdf.score.values
    idxs = non_max_suppression_poly(gdf.geometry.values, scores, nms_thresh, crit)
    gdf = gdf.iloc[idxs]
    return gdf

def do_min_rot_rectangle_nms(gdf:gpd.GeoDataFrame, nms_thresh=0.7, crit='score') -> gpd.GeoDataFrame:
    "Perform non-max suppression for rotated bounding boxes using `nms_threshold` to `gdf`"
    gdf = gdf.copy()
    scores = gdf.score.values
    boxes = np.array([g.minimum_rotated_rectangle for g in gdf.geometry.values])
    idxs = non_max_suppression_poly(boxes, scores, nms_thresh, crit)
    gdf = gdf.iloc[idxs]
    return gdf