训练模型以检测新对象 - YoloV3

Question

我使用this tutorial 训练我的模型来检测全息透镜。

现在，我创建了一个新的“吉他”Pascal Voc 数据集。

我想用检测全息透镜的实际模型来检测全息透镜和吉他。

在第 3 步中，我更改了以下几行：

trainer.setDataDirectory(data_directory="hololens")
trainer.setTrainConfig(object_names_array=["hololens"], batch_size=4, num_experiments=100, train_from_pretrained_model="pretrained-yolov3.h5")

到：

trainer.setDataDirectory(data_directory="violao")
trainer.setTrainConfig(object_names_array=["hololens", "violao"], batch_size=4, num_experiments=100, train_from_pretrained_model="drive/My Drive/PhoHast/my_model.h5")

“violao”文件夹结构是：

-violao
--train
---anottations
----<xml_files>
---images
----<images_files>

我在这些图片中的注释是“violao”。

当我运行这段代码时：

from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="violao")
trainer.setTrainConfig(object_names_array=["hololens", "violao"], batch_size=4, num_experiments=100, train_from_pretrained_model="my_model.h5")
trainer.trainModel()

我收到此错误：

Generating anchor boxes for training images and annotation...
[Errno 21] Is a directory: 'violao/train/annotations/.ipynb_checkpoints'
Ignore this bad annotation: violao/train/annotations/.ipynb_checkpoints
Average IOU for 9 anchors: 0.96
Anchor Boxes generated.
Detection configuration saved in  violao/json/detection_config.json
Some labels have no annotations! Please revise the list of labels in your configuration.
Training on:    None
Training with Batch Size:  4
Number of Experiments:  100
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-11-2e7a3bd0b6ee> in <module>()
      5 trainer.setDataDirectory(data_directory="violao")
      6 trainer.setTrainConfig(object_names_array=["hololens", "violao"], batch_size=4, num_experiments=100, train_from_pretrained_model="drive/My Drive/PhoHast/PhoHastV1.h5")
----> 7 trainer.trainModel()

1 frames
/usr/local/lib/python3.6/dist-packages/imageai/Detection/Custom/__init__.py in trainModel(self)
    231             shuffle=True,
    232             jitter=0.3,
--> 233             norm=normalize
    234         )
    235 

/usr/local/lib/python3.6/dist-packages/imageai/Detection/Custom/generator.py in __init__(self, instances, anchors, labels, downsample, max_box_per_image, batch_size, min_net_size, max_net_size, shuffle, jitter, norm)
     34         self.net_w              = 416
     35 
---> 36         if shuffle: np.random.shuffle(self.instances)
     37 
     38     def __len__(self):

mtrand.pyx in numpy.random.mtrand.RandomState.shuffle()

TypeError: object of type 'NoneType' has no len()

我做错了什么？？

Answer 1

我认为关键错误是这个

Some labels have no annotations! Please revise the list of labels in
your configuration.

我设法通过在 object_names_array 列表中添加训练数据集注释中不存在的新标签来重现类似问题，我得到了这些错误

Some labels have no annotations! Please revise the list of labels in your configuration.
...
Training on:    None
...
TypeError: object of type 'NoneType' has no len()

解决方案

1. 确保所有 object_names_array 标签都在您的数据集中，在您的情况下为“hololens”和“violao”。我假设您仅使用带有“violao”标记图像的数据集，因此 ImageAI 抱怨它无法找到 hololens 训练数据，即使您也指定在“hololens”上进行训练（假设您正在训练新模型并且它不知道训练前的任何标签，例如，如果你只留下 ["violao"] 它不会检测到 hololens，因为它对此一无所知。）
2. 从 data_directory 路径中删除不是 train 和 validation 的文件夹。当您添加标签等时，似乎从其他会话（很可能是缓存文件夹）生成的训练文件可能会干扰训练。

如果仍然不起作用，请尝试从给定的预训练 yolov3 模型 https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/pretrained-yolov3.h5 训练您的模型

Answer 2

试试这个：

trainer.setTrainConfig(object_names_array=[], batch_size=4, num_experiments=200, train_from_pretrained_model="---.h5")

object_names_array=[] 将默认从您的 xml 中选择注释。

Answer 3

我认为你用双 t 拼错了 annotations

我的问题是我在这里输入了文件夹的名称而不是对象

object_names_array=["object name in the Label"]