mask r-cnn 遇到的坑
mask r-cnn 遇到的坑 记录
BUG：module ‘tensorflow.compat.v2.internal‘ has no attribute ‘register_clear_session_function’ 已解决
更新tensorflow版本
tensorflow.python.keras.engine’ has no attribute ‘Layer’ 已解决
改为：
keras.layers.Layerhttps://www.e-learn.cn/topic/2433086
Tried to convert ‘shape’ to a tensor and failed. Error: None values not supported.
https://github.com/matterport/Mask_RCNN/issues/1820
NO.1
You can also try call tf.reshape(out_tensor, self.compute_output_shape(input_shape)) after each usage of Custom Layer.
More info:
https://stackoverflow.com/questions/51028861/tensorflow-compute-output-shape-not-working-for-custom-layer
tensorflow/tensorflow#33785
But in spite of this, the model summary differ in keras and tf.keras.
NO.2
To solve the problem, you can change this line:mrcnn_bbox = KL.Reshape((s[1], num_classes, 4), name="mrcnn_bbox")(x)
by :mrcnn_bbox = KL.Reshape((-1, num_classes, 4), name="mrcnn_bbox")(x)
It works for me and the output is fine.
tf1.13.2
解决编码问题：AttributeError: ‘str’ object has no attribute ‘decode’
- 出现问题原因：str与bytes表示的是两种数据类型，str为字符串型，bytes为字节型。对str编码encode得到bytes，
对bytes解码得到str，两者互为转换。而上面出现问题的原因是对str字符串使用了解码，显然是猪头不对马尾。 - 删除
decode('utf-8')
pip install h5py==2.10.0invalid literal for int() with base 10: ‘pths’
nameinfo = "pths"
image_info.update({"id": int(name_info)})强制转换成int类型时，传入的字符串必须是有数字字符组成的，例如：x = int("111")
加载权重时出现“AttributeError: ‘str‘ object has no attribute ‘decode‘ “
解决办法
貌似是由于h5py模块的版本问题，改用2.10版本
pip install h5py==2.10 -i https://pypi.doubanio.com/simple发生异常: IndexError ：boolean index did not match indexed array along dimension 0; dimension is 0 but corresponding boolean dimension is 1
暂未解决，环境配置成一模一样也没用
后来发现没有更改分类的参数T_T
Input image dtype is bool. Interpolation is not defined with bool data type
pip install -U scikit-image==0.16.2Error when checking input: expected input_image_meta to have shape (15,) but got array with shape (16,)
解决：因为我增加了一类，但是没有修改class_nums
解决办法是修改NUM_CLASSES参数，增加了多少类，就对应的修改对应数值
NUM_CLASSES = 1 + 1 https://blog.csdn.net/oMoDao1/article/details/83860101
AttributeError: module ‘yaml’ has no attribute ‘FullLoader’
pip install --ignore-installed PyYAML[1].AttributeError: ‘module’ object has no attribute ‘FullLoader’. https://github.com/Yelp/elastalert/issues/2298
[2].安装yaml报错：ERROR: Cannot uninstall ‘PyYAML’. https://blog.csdn.net/weixin_41010198/article/details/103852838
keras 导入包报错 ImportError: cannot import name ‘moving_averages’
pip3 install --upgrade --ignore-installed tensorflow-gpu==1.13.2https://github.com/tensorflow/tensorflow/issues/8513
AttributeError: module ‘labelme.utils’ has no attribute ‘draw_label’
将版本降为3.16.2
pip install labelme==3.16.2IndexError: boolean index did not match indexed array along dimension 0; dimension is 0 but corresponding boolean dimension is 1
新版本的labelme生成的label.png图为8位彩色图，不需要16位转8位转换！！！直接使用即可！注意观察自己的label.png的位深度！
BUG
---------------------------------------------------------------------------
AssertionError Traceback (most recent call last)
<ipython-input-28-90029a2660a7> in <module>()
13 ("refined_anchors_clipped", model.ancestor(pillar, "ROI/refined_anchors_clipped:0")),
14 ("post_nms_anchor_ix", nms_node),
---> 15 ("proposals", model.keras_model.get_layer("ROI").output),
16 ])
/content/drive/My Drive/mask rcnn_test/mrcnn/model.py in run_graph(self, images, outputs, image_metas)
2663 outputs = OrderedDict(outputs)
2664 for o in outputs.values():
-> 2665 assert o is not None
2666
2667 # Build a Keras function to run parts of the computation graph
AssertionError: 暂未找到解决方法，或许：
- 版本问题，使用的是1.13 1.15
- 是分割单类别，多类别是否会出现？
wechat
alipay
