尝试使用 detector 训练对象检测模型并在数据集中出现 CV2 错误

Question

我正在使用 detector 和 Google Colab Notebook 来训练和测试我的数据集。我使用 google 和 labelImg 的图像下载器从我的图像创建 .xml 文件。我将它们存储在我的驱动器中的 object_detection 文件夹中。我收到以下错误：

Epoch 1 of 25
Begin iterating over training dataset
 68%|██████▊   | 75/111 [02:03<00:59,  1.64s/it]
---------------------------------------------------------------------------
error                                     Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/detecto/utils.py in read_image(path)
    137     try:
--> 138         rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    139     except cv2.error as e:

error: OpenCV(4.1.2) /io/opencv/modules/imgproc/src/color.cpp:182: error: (-215:Assertion failed) !_src.empty() in function 'cvtColor'


During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
8 frames
/usr/local/lib/python3.7/dist-packages/detecto/utils.py in read_image(path)
    138         rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    139     except cv2.error as e:
--> 140         raise ValueError(f'Could not convert image color: {str(e)}')
    141 
    142     return rgb_image

ValueError: Could not convert image color: OpenCV(4.1.2) /io/opencv/modules/imgproc/src/color.cpp:182: error: (-215:Assertion failed) !_src.empty() in function 'cvtColor'

这是我当前拥有的代码：_________________________

import torch
print(torch.cuda.is_available())
print(torch.cuda.get_device_name(0))

from detecto import core, utils, visualize
import numpy as np
import seaborn as sns
import torchvision.transforms as tf
from detecto.visualize import show_labeled_image, plot_prediction_grid

custom_transforms = tf.Compose([
tf.ToPILImage(),
tf.Resize(900),
tf.RandomHorizontalFlip(0.5),
tf.ColorJitter(saturation=0.2),
tf.ToTensor(),
utils.normalize_transform(),
])

dataset = core.Dataset('images/', transform = custom_transforms)
loader=core.DataLoader(dataset, batch_size=2, shuffle=True)
model = core.Model(['Card', 'Card Name'])
losses = model.fit(loader, dataset, epochs=25, lr_step_size=5, learning_rate= 0.001, verbose=True) 

---

或者，我也尝试过：

import torch
print(torch.cuda.is_available())
print(torch.cuda.get_device_name(0))

from detecto import core, utils, visualize
import numpy as np
from detecto.visualize import show_labeled_image, plot_prediction_grid

dataset = core.Dataset('images/')
model = core.Model(['Card', 'Card Name'])
model.fit(dataset)

请帮助我解决此问题。

Answer 1

我尝试使用包含 30 张图像的较小数据集来训练模型，结果发现 .gif 图像意外出现在我的数据集中并且无法读取。确保数据集中没有 .gif 文件，如果出现此错误，请使用较小的测试集。