[Python] Why don't I get the expected neighbor indices with OpenCVs KNearestNeighbour?

I'm trying to use OpenCVs KNearestNeighbours in python. I'd like to use it to extract Euclidean distances to the nearest neighbour. The outputs are not what I expect, which I think means I'm either incorrect about what the function is supposed to return OR I'm inputting incorrect arguments.

Specifically I'd like to know how to correctly get the neighbour indices

Here's an example where I compute distances both manually and with OpenCV. Distances match but indices do not.

Update


I originally asked about distances as well but now realize that opencv returns squared distances. Code and output are adjusted accordingly.

Code


import numpy as np
import cv2


np.random.seed(42)

# Example training data with two features
X_train = np.random.rand(25, 2).astype(np.float32) # 25 samples, 2 features
y_train = np.random.randint(0, 2, (25, 1)).astype(np.float32) # 25 labels (binary classification)

# Example test data with two features
X_test = np.random.rand(5, 2).astype(np.float32) # 5 test samples, 2 features

# Create and train the k-NN model
knn = cv2.ml.KNearest_create()
knn.train(X_train, cv2.ml.ROW_SAMPLE, y_train)

# Perform k-NN classification
ret, results, neighbors, distances = knn.findNearest(X_test, k=1)

# Manually calculate Euclidean distances
for i, test_sample in enumerate(X_test):
# Calculate Euclidean distances to all training samples
manual_distances = np.sum((X_train - test_sample) ** 2, axis=1)

# Find the nearest neighbor manually
nearest_index = np.argmin(manual_distances)
nearest_distance = manual_distances[nearest_index]

# Compare with OpenCV's output
print(f"Test sample {i}:")
print(f" Manual nearest distance: {nearest_distance}")
print(f" OpenCV nearest distance: {distances[0]}")
print(f" Manual nearest neighbor index: {nearest_index}")
print(f" OpenCV nearest neighbor index: {neighbors[0]}")

Output


Test sample 0:
Manual nearest distance: 0.0452449657022953
OpenCV nearest distance: 0.0452449657022953
Manual nearest neighbor index: 0
OpenCV nearest neighbor index: 0.0

Test sample 1:
Manual nearest distance: 0.011125250719487667
OpenCV nearest distance: 0.011125250719487667
Manual nearest neighbor index: 15
OpenCV nearest neighbor index: 1.0

Test sample 2:
Manual nearest distance: 0.004638621583580971
OpenCV nearest distance: 0.004638621583580971
Manual nearest neighbor index: 18
OpenCV nearest neighbor index: 1.0

Test sample 3:
Manual nearest distance: 0.016473062336444855
OpenCV nearest distance: 0.016473062336444855
Manual nearest neighbor index: 20
OpenCV nearest neighbor index: 1.0

Test sample 4:
Manual nearest distance: 0.0013788070064038038
OpenCV nearest distance: 0.0013788070064038038
Manual nearest neighbor index: 9
OpenCV nearest neighbor index: 0.0

Continue reading...