1.Numpy的几个骚操作
(1).np.maximum的使用
import numpy as np
box = [3,5,7,9] # A single box with a first coordinate of 3
boxes = np.array([[1, 4],
[5, 2],
[2, 6]]) # An array of multiple boxes
#把box把3拿出来和boxes的第一列，逐个比较，返回最大值list
result = np.maximum(box[0], boxes[:, 0])
#result=[3 5 3]
print(result)
(2). np.argsort的使用：
import numpy as np
scores = np.array([0.5, 0.2, 0.8])
list1 = np.argsort(scores)
list2=np.argsort(scores)[::-1]
#list1=[1 0 2]
print(f"list1={list1}")
#list2=[2 0 1]
print(f"list2={list2}")
#scores=[0.5 0.2 0.8],原来的顺序没改
print(f"scores={scores}")
(3).向量和矩阵的乘法和加法
import numpy as np
box = [3,5] # A single box with a first coordinate of 3
boxes = np.array([[1, 4],
[5, 2],
[2, 6]]) # An array of multiple boxes
"""
矩阵乘法是"@"不是"*"
*不是矩阵乘法，把boxes的每一行都对应乘以box
box*boxes=[[ 3 20]
[15 10]
[ 6 30]]
"""
print(box*boxes)
"""
boxes的每一列都对应乘以box
box+boxes=[[ 4 9]
[ 8 7]
[ 5 11]]
"""
print(box+boxes)
2.非极大值抑制
def non_max_suppression(boxes, scores, threshold):
"""
Apply non-maximum suppression to eliminate redundant bounding boxes.
Parameters:
- boxes: A list of bounding boxes in the format (x1, y1, x2, y2).
- scores: A list of confidence scores for each bounding box.
- threshold: The IoU (Intersection over Union) threshold for suppression.
Returns:
- selected_indices: A list of indices for the selected bounding boxes.
"""
if len(boxes) == 0:
return []
# Convert the bounding boxes to the format (x1, y1, x2, y2)
boxes = np.array(boxes)
# Sort indices based on confidence scores (in descending order)
#sorted_indices=[2,0,1]
sorted_indices = np.argsort(scores)[::-1]
selected_indices = []
while len(sorted_indices) > 0:
# Pick the bounding box with the highest confidence score
i = sorted_indices[0]
selected_indices.append(i)
# Calculate IoU with the remaining bounding boxes
ious = calculate_iou(boxes[i], boxes[sorted_indices[1:]])
# Filter out bounding boxes with IoU higher than the threshold
#np.where(ious <= threshold) is ([])
#type(remaining_indices)=ndarray
remaining_indices = np.where(ious <= threshold)[0]
print(f"remaining_indices={remaining_indices}")
print(f"remaining_indices+1={remaining_indices+1}")
#i=2的哪个框已经进入selected
sorted_indices = sorted_indices[remaining_indices + 1]
print("sorted_indices=",sorted_indices)
return selected_indices
def calculate_iou(box, boxes):
"""
Calculate the Intersection over Union (IoU) between a bounding box and a list of bounding boxes.
Parameters:
- box: The bounding box in the format (x1, y1, x2, y2).
- boxes: A list of bounding boxes.
Returns:
- ious: A numpy array of IoU values.
"""
#计算intersection的4个坐标
x1 = np.maximum(box[0], boxes[:, 0])
y1 = np.maximum(box[1], boxes[:, 1])
x2 = np.minimum(box[2], boxes[:, 2])
y2 = np.minimum(box[3], boxes[:, 3])
#没有交集的框的IoU就是0
intersection = np.maximum(0, x2 - x1) * np.maximum(0, y2 - y1)
area_box = (box[2] - box[0]) * (box[3] - box[1])
area_boxes = (boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1])
union = area_box + area_boxes - intersection
ious = intersection / union
return ious
# Example usage:
import numpy as np
if __name__=="__main__":
# Example bounding boxes and confidence scores
#the 4 number in each box is the coordinate of left-top corner
#and right-bottom corner
boxes = [(10, 20, 50, 60), (15, 25, 55, 65), (30, 40, 70, 80)]
scores = [0.9, 0.75, 0.95]
# Set the IoU threshold for non-maximum suppression
iou_threshold = 0.5
# Apply non-maximum suppression
selected_indices = non_max_suppression(boxes, scores, iou_threshold)
# Display the selected bounding boxes
for i in selected_indices:
print(f"Selected Box: {boxes[i]}, Score: {scores[i]}")
Selected Box: (30, 40, 70, 80), Score: 0.95
Selected Box: (10, 20, 50, 60), Score: 0.9
