Why am I facing error in object detection?

Question

confidence_score = scores[class1]

IndexError: index 172 is out of bounds for axis 0 with size 5

import cv2
import numpy as np
net = cv2.dnn.readNet("yolov3.weights", "yolov3.cfg.txt")
classes = []
img1 = cv2.imread('img1.jpg')
img1 = cv2.resize(img1,None, fx =0.4, fy =0.4)
height,width,chanels = img1.shape
with open("coco.names.txt", "r") as f:
     classes = [line.strip() for line in f.readlines()]
layer_name = net.getLayerNames()
output_layers = [layer_name[i[0] - 1] for i in net.getUnconnectedOutLayers()]
floaty =0.004
blob = cv2.dnn.blobFromImage(img1,floaty,(416,416),(0,0,0),True)
# true to convert RBG
for b in blob:
    for n,img_blog in enumerate(b):
        cv2.imshow(str(n), img_blog)
net.setInput(blob)
out = net.forward(output_layers)

#trying to show or detect
for show in out:
    for detection in out:
        scores = detection[:5]
        class1 = np.argmax(scores)
        confidence_score = scores[class1]
        if confidence_score > 0.6:
            center_x = int[detection[0] * width]
            center_y = int(detection[1] * height)
            w = int(detection[2] * width)
            h = int(detection[3] * height)
            cv2.circle(img1,(center_x,center_y),12,(0,255,0),2)

cv2.imshow('image', img1)
cv2.waitKey(0)
cv2.destroyAllWindows()

the error is confidence_score = scores[class1] IndexError: index 172 is out of bounds for axis 0 with size 5 — Madhav Bajaj, Jul 11 '20 at 17:21
Does this answer your question? [Argmax of numpy array returning non-flat indices](https://stackoverflow.com/questions/9482550/argmax-of-numpy-array-returning-non-flat-indices) — adamgy, Jul 11 '20 at 17:31
confidence_score = scores[class1] IndexError: index 172 is out of bounds for axis 0 with size 5 — Madhav Bajaj, Jul 11 '20 at 17:36

adamgy · Answer 1 · 2020-07-11T18:09:09.160

1

The problem is that using numpy.argmax without the axis argument returns the index as if the array were flattened:

a = np.arange(6).reshape(2,3) + 10
# array([[10, 11, 12],
#       [13, 14, 15]])

flat_max = np.argmax(a)
# 5 <=== "flattened" indices of max element (15)

a[flat_max]
# IndexError: index 5 is out of bounds for axis 0 with size 2

You can use np.unravel_index() on the result of numpy.argmax() to get the correct indices of the maximum element:

i, j = np.unravel_index(np.argmax(a), a.shape)
a[i][j]
# (1,2) <=== correct indices of max element (15)

In your code, you get 172 as the result of np.argmax(scores), but since axis zero of scores only has a size of 5, you get an IndexError. Use np.unravel_index() as shown above and use the returned indices to index into scores:

i, j = np.unravel_index(np.argmax(scores), scores.shape)
confidence_score = scores[i][j]

edited Jul 11 '20 at 18:09

answered Jul 11 '20 at 17:51

adamgy

4,543
3
16
31

can you give code as per my question because I was following a tutorial and I do not have much knowledge @adamgy – Madhav Bajaj Jul 11 '20 at 17:54
what are the dimensions of score? (you can check by inserting print(scores.shape) in your code) – adamgy Jul 11 '20 at 18:01
where shall i print the error is only getting thrown – Madhav Bajaj Jul 11 '20 at 18:05
ok, try replacing your two lines of code starting from class1 = ... with the two lines I just added to my answer – adamgy Jul 11 '20 at 18:10
now this error is coming " TypeError: only size-1 arrays can be converted to Python scalars Line 25" – Madhav Bajaj Jul 11 '20 at 18:17
are you sure the shape of scores is (502, 85)? Your first error message suggests that axis zero has a size of 5 – adamgy Jul 11 '20 at 18:55

Why am I facing error in object detection?

1 Answers1