I was reading through the paper : Ferrari et al. in the "Affinity Measures" section. I understood that Ferrari et al. tries to obtain affinity by :
- Location affinity - using area of intersection-over-union between two detections
- Appearance affinity - using Euclidean distances between Histograms
- KLT point affinity measure
However, I have 2 main problems:
- I cannot understand what is actually meant by intersection-over-union between 2 detections and how to calculate it
- I tried a slightly difference appearance affinity measure. I transformed the RGB detection into HSV..concatenating the Hue and Saturation into 1 vector, and used it to compare with other detections. However, using this technique failed as a detection of a bag had a better similarity score than a detection of the same person's head (with a different orientation).
Any suggestions or solutions to my problems described above? Thank you and your help is very much appreciated.
1) You have two overlapping bounding boxes. You compute the intersection of the boxes, which is the area of the overlap. You compute the union of the overlapping boxes, which is the sum of the areas of the entire boxes minus the area of the overlap. Then you divide the intersection by the union. There is a function for that in the Computer Vision System Toolbox called bboxOverlapRatio.
2) Generally, you don't want to concatenate the color channels. What you want instead, is a 3D histogram, where the dimensions are H, S, and V.
Try intersection over Union
Intersection over Union is an evaluation metric used to measure the accuracy of an object detector on a particular dataset.
More formally, in order to apply Intersection over Union to evaluate an (arbitrary) object detector we need:
- The ground-truth bounding boxes (i.e., the hand labeled bounding boxes from the testing set that specify where in the image our object is).
- The predicted bounding boxes from our model.
Below I have included a visual example of a ground-truth bounding box versus a predicted bounding box:
The predicted bounding box is drawn in red while the ground-truth (i.e., hand labeled) bounding box is drawn in green.
In the figure above we can see that our object detector has detected the presence of a stop sign in an image.
Computing Intersection over Union can therefore be determined via:
As long as we have these two sets of bounding boxes we can apply Intersection over Union.
Here is the Python code
# import the necessary packages from collections import namedtuple import numpy as np import cv2 # define the `Detection` object Detection = namedtuple("Detection", ["image_path", "gt", "pred"]) def bb_intersection_over_union(boxA, boxB): # determine the (x, y)-coordinates of the intersection rectangle xA = max(boxA[0], boxB[0]) yA = max(boxA[1], boxB[1]) xB = min(boxA[2], boxB[2]) yB = min(boxA[3], boxB[3]) # compute the area of intersection rectangle interArea = (xB - xA) * (yB - yA) # compute the area of both the prediction and ground-truth # rectangles boxAArea = (boxA[2] - boxA[0]) * (boxA[3] - boxA[1]) boxBArea = (boxB[2] - boxB[0]) * (boxB[3] - boxB[1]) # compute the intersection over union by taking the intersection # area and dividing it by the sum of prediction + ground-truth # areas - the interesection area iou = interArea / float(boxAArea + boxBArea - interArea) # return the intersection over union value return iou
The gt
and pred
are
gt
: The ground-truth bounding box. pred
: The predicted bounding box from our model.
For more information, you can click this post