/// <summary>
/// Evaluates the similarity between two maps using the graph sampling evaluation of biagioni and eriksson.
/// Parameters are hardcoded, based on the ones we used in the corresponding research.
/// </summary>
/// <param name="GT">The ground truth map</param>
/// <param name="CT">the constructed map.</param>
/// <param name="amount">the amount of neighbourhood evaluations we average over</param>
/// <returns>A precision and recall value indicating the similarity between the maps.</returns>
public (float, float) EvalMap(Map GT, Map CT, int amount)
{
float precision = 0;
float recall = 0;
for (int k = 0; k < amount; k++)
{
Coordinate originGT = rand.GetRandomPointOnRoad(rand.GetWeightedRandomRoad(GT));
Coordinate originCT = CT.GetClosestPoint(originGT.location);
/* An extra step that is sometimes used in the literature when ground truth map is not pruned.
* while (Vector3.Distance(originCT.location, originGT.location) > 50)
* {
* originGT = rand.GetRandomCoordinate(rand.GetWeightedRandomRoad(GT));
* originCT = CT.GetClosestPoint(originGT.location);
* }
*/
List <Vector3> pointsGT = sampleNeighbourhood.GetNeighbourhood(GT, originGT, 500, 30);
List <Vector3> pointsCT = sampleNeighbourhood.GetNeighbourhood(CT, originCT, 500, 30);
MaxFlow flow = new MaxFlow(pointsGT.Count + pointsCT.Count + 2);
for (int i = 0; i < pointsCT.Count; i++)
{
flow.AddEdge(0, i + 1, 1);
}
for (int i = 0; i < pointsCT.Count; i++)
{
Vector3 pointCT = pointsCT[i];
for (int j = 0; j < pointsGT.Count; j++)
{
Vector3 pointGT = pointsGT[j];
if (Vector3.Distance(pointCT, pointGT) < 20)
{
flow.AddEdge(i + 1, j + pointsCT.Count + 1, 1);
}
}
}
for (int i = 0; i < pointsGT.Count; i++)
{
flow.AddEdge(i + pointsCT.Count + 1, 1 + pointsCT.Count + pointsGT.Count, 1);
}
float matching = flow.FindMaximumFlow(0, 1 + pointsCT.Count + pointsGT.Count).Item1;
precision += pointsCT.Count > 0 ? matching / pointsCT.Count : 0;
recall += pointsGT.Count > 0 ? matching / pointsGT.Count : 0;
}
return(precision / amount, recall / amount);
}
/// <summary>
/// Visualizes and evaluates a single random neighbourhood.
/// </summary>
/// <param name="GT">Ground truth map</param>
/// <param name="CT">contructed map</param>
/// <param name="originDistanceCondition"></param>
/// <param name="matchDistance"></param>
/// <returns></returns>
public (float, float) EvalNeighbourhood(Map GT, Map CT)
{
Coordinate originGT = rand.GetRandomPointOnRoad(rand.GetWeightedRandomRoad(GT));
Coordinate originCT = CT.GetClosestPoint(originGT.location);
/* An extra step that is sometimes used in the literature when ground truth map is not pruned.
* while (Vector3.Distance(originCT.location, originGT.location) > 50)
* {
* originGT = rand.GetRandomCoordinate(rand.GetWeightedRandomRoad(GT));
* originCT = CT.GetClosestPoint(originGT.location);
* }
*/
List <Vector3> pointsGT = sampleNeighbourhood.GetNeighbourhood(GT, originGT, 150, 30);
List <Vector3> pointsCT = sampleNeighbourhood.GetNeighbourhood(CT, originCT, 150, 30);
MaxFlow flow = new MaxFlow(pointsGT.Count + pointsCT.Count + 2);
for (int i = 0; i < pointsCT.Count; i++)
{
flow.AddEdge(0, i + 1, 1); // source edges
}
for (int i = 0; i < pointsCT.Count; i++) // edges between CT and GT
{
Vector3 pointCT = pointsCT[i];
for (int j = 0; j < pointsGT.Count; j++)
{
Vector3 pointGT = pointsGT[j];
if (Vector3.Distance(pointCT, pointGT) < 20)
{
flow.AddEdge(i + 1, j + pointsCT.Count + 1, 1);
}
}
}
for (int i = 0; i < pointsGT.Count; i++)
{
flow.AddEdge(i + pointsCT.Count + 1, 1 + pointsCT.Count + pointsGT.Count, 1); // sink edges
}
float matching;
int[,] graph;
(matching, graph) = flow.FindMaximumFlow(0, 1 + pointsCT.Count + pointsGT.Count);
float prec = pointsCT.Count > 0 ? matching / pointsCT.Count : 0;
float recall = pointsGT.Count > 0 ? matching / pointsGT.Count : 0;
return(prec, recall);
}
