MachineLearning
1473684220@ak11

IoUの0.1～0.9を図にしてみた

MachineLearning

はじめに

物体検出とかではIoU(Intersection over Union、Jaccard index)をよく使いますが、どのくらいの数値でどのくらいの重なり具合なのかがよくわからなかったので図にしてみました。

図

IoU == 0.9

#1 iou=0.90, offset=0.03.png

IoU == 0.8

#2 iou=0.80, offset=0.06.png

IoU == 0.7

#3 iou=0.70, offset=0.09.png

IoU == 0.6

#4 iou=0.60, offset=0.13.png

IoU == 0.5

#5 iou=0.50, offset=0.18.png

IoU == 0.4

#6 iou=0.40, offset=0.24.png

IoU == 0.3

#7 iou=0.30, offset=0.32.png

IoU == 0.2

#8 iou=0.20, offset=0.42.png

IoU == 0.1

#9 iou=0.10, offset=0.57.png

コード

plot_iou.py
import matplotlib
import matplotlib.pyplot as plt

import numpy as np

for i, iou in enumerate(np.linspace(1, 0, 11)):
    # 　 ↓t
    # □□□
    # □□□
    # □□■□□
    # 　　□□□
    # 　　□□□

    # inter = (1 - t) ** 2
    # union = 2 - inter
    # iou = inter / union
    # https://www.wolframalpha.com/input/?dataset=&i=Solve%5Bu+%3D%3D+(1+-+t)%5E2+%2F+(2+-+(1+-+t)%5E2)+%26%26+0+<%3D+t+%26%26+t+<%3D+1+%26%26+0+<%3D+u+%26%26+u+<%3D+1,+t%5D
    # → t = 1 - sqrt(2) sqrt(u/(1 + u))
    t = 1 - np.sqrt(2 * iou / (1 + iou))
    print(f'#{i} iou={iou:.2f}, offset={t:.2f}')

    fig = plt.figure()
    ax = plt.axes()
    ax.add_patch(matplotlib.patches.Rectangle(xy=(0 / 2, 0.5 + 0 / 2), width=0.5, height=0.5, color='#9999ff77'))
    ax.add_patch(matplotlib.patches.Rectangle(xy=(t / 2, 0.5 - t / 2), width=0.5, height=0.5, color='#99ff9977'))
    ax.set_aspect('equal')
    plt.savefig(f'#{i} iou={iou:.2f}, offset={t:.2f}.png')
    plt.close()
Edit request
8