我现在正在编写导入 Keras 的对象检测 python 代码 这是一组形状为(175,480,720,3)的图像，图像数量为175。 还有一组形状为 (175, 4) 的标签与图像对应，图像中的每个对象都对应一个标签。边界框 (175, 4, 4) 是相同的。
model.fit(trainImages, {"classification": trainLabels, "regression": trainBboxes},
validation_data=(validImages, {"classification": validLabels, "regression": validBboxes}),
batch_size=8, epochs=10)
我将它们输入到我的神经网络中，但它显示错误
ValueError：数据基数不明确。确保所有数组包含相同数量的样本。“x”大小：480, 480, ... 480, 480(175) “y”尺寸：4, 4, ...4, 4(175)
我将 np 数组转换为 list ，但什么也没发生。 完整代码如下：
import os
import numpy as np
import xml.etree.ElementTree as ET
import cv2
import tensorflow as tf
from keras import layers
import keras
print("模块导入完成")
def get_annotations(cname2cid, datadir):
filenames = os.listdir(os.path.join(datadir, 'annotations', 'xml'))
records = []
ct = 0
for fname in filenames:
fid = fname.split('.')[0]
fpath = os.path.join(datadir, 'annotations', 'xml', fname)
img_file = os.path.join(datadir, 'images', fid + '.png')
tree = ET.parse(fpath)
if tree.find('id') is None:
im_id = np.array([ct])
else:
im_id = np.array([int(tree.find('id').text)])
objs = tree.findall('object')
im_w = float(tree.find('size').find('width').text)
im_h = float(tree.find('size').find('height').text)
gt_bbox = np.zeros((len(objs), 4), dtype=np.float32)
gt_class = np.zeros((len(objs), ), dtype=np.int32)
is_crowd = np.zeros((len(objs), ), dtype=np.int32)
difficult = np.zeros((len(objs), ), dtype=np.int32)
for i, obj in enumerate(objs):
cname = obj.find('name').text
gt_class[i] = cname2cid[cname]
_difficult = int(obj.find('difficult').text)
x1 = float(obj.find('bndbox').find('xmin').text)
y1 = float(obj.find('bndbox').find('ymin').text)
x2 = float(obj.find('bndbox').find('xmax').text)
y2 = float(obj.find('bndbox').find('ymax').text)
x1 = max(0, x1)
y1 = max(0, y1)
x2 = min(im_w - 1, x2)
y2 = min(im_h - 1, y2)
# 这里使用xywh格式来表示目标物体真实框
gt_bbox[i] = [(x1+x2)/2.0, (y1+y2)/2.0, x2-x1+1., y2-y1+1.]
is_crowd[i] = 0
difficult[i] = _difficult
voc_rec = {
'im_file': img_file,
'im_id': im_id,
'h': im_h,
'w': im_w,
'is_crowd': is_crowd,
'gt_class': gt_class,
'gt_bbox': gt_bbox,
'gt_poly': [],
'difficult': difficult
}
if len(objs) != 0:
records.append(voc_rec)
ct += 1
return records
def create_ssd(input_shape):
# 输入层
inputs = keras.Input(shape=input_shape)
# VGG16 主干网络
base_model = keras.applications.VGG16(
weights="imagenet", include_top=False, input_tensor=inputs)
# 添加额外的卷积层和池化层
x = base_model.output
x = layers.Conv2D(256, 3, padding="same", activation="relu")(x)
x = layers.Conv2D(256, 3, padding="same", activation="relu")(x)
x = layers.Conv2D(128, 3, padding="same", activation="relu")(x)
x = layers.Conv2D(128, 3, padding="same", activation="relu")(x)
# 添加检测头
num_classes = 2
num_boxes = 4
classification_outputs = layers.Conv2D(num_classes, 3, padding="same")(x)
classification_outputs = layers.Reshape(
(-1, num_classes))(classification_outputs)
classification_outputs = layers.Activation(
"softmax", name="classification")(classification_outputs)
regression_outputs = layers.Conv2D(num_boxes * 4, 3, padding="same")(x)
regression_outputs = layers.Reshape((-1, num_boxes, 4))(regression_outputs)
regression_outputs = layers.Activation(
"linear", name="regression")(regression_outputs)
# 创建模型
model = keras.Model(inputs=inputs, outputs=[
classification_outputs, regression_outputs], name="SSD")
return model
def drawRec(num):
pic = trainImages[num]
colorDict = {0:(0,0,255),1:(255,0,0)}
colors = [colorDict[i] for i in list(trainRecords[num]["gt_class"])]
x = 0
for gt_bbox in list(trainRecords[num]["gt_bbox"]):
color = colors[x]
x+=1
cv2.rectangle(pic,(int(gt_bbox[0]-gt_bbox[2]/2), int(gt_bbox[1]-gt_bbox[3]/2)),
(int(gt_bbox[0]+gt_bbox[2]/2), int(gt_bbox[1]+gt_bbox[3]/2)),color,2)
return pic
if __name__ == "__main__":
cname = {"book": 0, "bottle": 1}
trainRecords = get_annotations(cname, datadir="E:\\ret\\train")
validRecords = get_annotations(cname, datadir="E:\\ret\\valid")
trainImages = [cv2.imread(record["im_file"]) for record in trainRecords]
validImages = [cv2.imread(record["im_file"]) for record in validRecords]
trainLabels = [dlabel["gt_class"] for dlabel in trainRecords]
validLabels = [dlabel["gt_class"] for dlabel in validRecords]
trainBboxes = [dbbox["gt_bbox"] for dbbox in trainRecords]
validBboxes = [dbbox["gt_bbox"] for dbbox in validRecords]
# check
"""
num = np.random.randint(0,174)
pic = drawRec(num)
while True:
cv2.imshow("pic",pic)
key = cv2.waitKey(1)
if key & 0xFF == ord('q'):
break
elif (key & 0xFF == ord('.')) and (num<174):
num+=1
pic = drawRec(num)
elif (key & 0xFF == ord(',')) and (num>-175):
num -= 1
pic = drawRec(num)
cv2.destroyAllWindows()
trainImages = np.array(trainImages)
validImages = np.array(validImages)
print(np.shape(trainImages))
print(trainImages[num])
print(trainLabels[num])
print(trainBboxes[num])
"""
for i in range(len(trainLabels)):
while len(trainLabels[i]) <4:
trainLabels[i] = np.append(trainLabels[i],-1)
for i in range(len(trainBboxes)):
while len(trainBboxes[i]) <4:
x = list(trainBboxes[i])
x.append(np.array([0.,0.,0.,0.]))
trainBboxes[i] = np.array(x)
for i in range(len(validLabels)):
while len(validLabels[i]) <4:
validLabels[i] = np.append(validLabels[i],-1)
for i in range(len(validBboxes)):
while len(validBboxes[i]) <4:
x = list(validBboxes[i])
x.append(np.array([0.,0.,0.,0.]))
validBboxes[i] = np.array(x)
print(np.shape(trainLabels))
print(np.shape(trainImages))
# print(trainLabels[num])
print(np.shape(trainBboxes))
# print(trainBboxes[num])
model = create_ssd((480, 720, 3))
model.compile(optimizer=keras.optimizers.Adam(learning_rate=1e-3),
loss={"classification": "categorical_crossentropy",
"regression": "mse"},
metrics={"classification": "accuracy"})
model.fit(trainImages, {"classification": trainLabels, "regression": trainBboxes},
validation_data=(validImages, {"classification": validLabels, "regression": validBboxes}),
batch_size=8, epochs=10)
model.save_weights(filepath="model.h5", save_format="h5")
您似乎正在尝试使用 Keras 训练用于对象检测的模型。您提供的代码片段演示了如何使用 fit() 函数来训练具有分类和回归任务的模型。 以下是传递给 fit() 函数的参数的详细信息：