Commit 74e53e3f authored by kberry's avatar kberry
Browse files

Added adaptive_non_max_suppression to outlier_detector.py

parent c4a50ff6

autocnet/matcher/outlier_detector.py

+25 −0
Original line number Diff line number Diff line 
import cv2

import numpy as np

import pandas as pd





def self_neighbors(matches):
@@ -150,3 +151,27 @@ def compute_homography(kp1, kp2, outlier_algorithm=cv2.RANSAC, reproj_threshold= 
    return transformation_matrix, mask





# TODO: document me

# right now, this is going to work on the un-merged dataframes... NOPE. I confused what was where again. Need the

# graph for keypoint information...

def adaptive_non_max_suppression(keypoints, n=20, robust=0.9):

    minimum_suppression_radius = {}

    for i, kp1 in enumerate(keypoints):

        x1, y1 = kp1.pt

        temp = []

        for kp2 in keypoints: #includes kp1 for now

            if kp1.response < robust*kp2.response:

                x2, y2 = kp2.pt

                temp.append(np.sqrt((x2-x1)**2 + (y2-y1)**2))

        if(len(temp) > 0):

            minimum_suppression_radius[i] = np.min(np.array(temp))

        else:

            minimum_suppression_radius[i] = [np.nan]

    df = pd.DataFrame(list(minimum_suppression_radius.items()), columns=['keypoint_number', 'radius'])

    new_df = df.sort_values(by='radius', ascending=False).head(n)

    return new_df

functional_tests/test_two_image.py

+5 −0
Original line number Diff line number Diff line
@@ -49,6 +49,11 @@ class TestTwoImageMatching(unittest.TestCase): 
        for node, attributes in cg.nodes_iter(data=True):

            self.assertIn(len(attributes['keypoints']), range(490, 511))



        # Step: apply ANMS

        for node, attributes in cg.nodes_iter(data=True):

            print(od.adaptive_non_max_suppression(attributes['keypoints']))

        self.assertTrue(False)



        # Step: Then apply a FLANN matcher

        fl = FlannMatcher()

        for node, attributes in cg.nodes_iter(data=True):