Code in tutorial throws AttributeError: 'FigureCanvasTkAgg' object has no attribute 'tostring_rgb'. Did you mean: 'tostring_argb'?

[the-big-bad-wolf]

I tried running the code in the Scenario Data Loading tutorial, but the program crashed.

Traceback (most recent call last):
File "c:\Users----\Documents\WaymaxAutoPlanner\main.py", line 16, in
img = visualization.plot_simulator_state(scenario, use_log_traj=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users----\Documents\WaymaxAutoPlanner.venv\Lib\site-packages\waymax\visualization\viz.py", line 311, in plot_simulator_state
return utils.img_from_fig(fig)
^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users----\Documents\WaymaxAutoPlanner.venv\Lib\site-packages\waymax\visualization\utils.py", line 89, in img_from_fig
data = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'FigureCanvasTkAgg' object has no attribute 'tostring_rgb'. Did you mean: 'tostring_argb'?

[vcharraut]

Downgrading matplotlib to a version below 3.10 fix the issue

pip install matplotlib<3.10

[zuodong-zhong]

ValueError: cannot reshape array of size 1440000 into shape (600,600,3)

出错文件：waymax.utils.img_from_fig()

这是因为，waymax的utils.img_from_fig(fig)函数中，在处理图像时，先转成了RGB的一维数组形式，然后再重塑成图像的三维数组（h,w,r,g,b）格式，但我图像转换的函数库版本高，已经弃用了RGB的方法（所以我就换成了ARGB的格式，也就导致了上面的不匹配的错误），使用的是ARGB的格式（就是加了一个透明度），就无法适配指定的（h,w,r,g,b）格式了，所以我就把（h,w,r,g,b）格式改成了（h,w,r,g,b,a）格式，就画出来了，图片上应该是加上了透明度，修改后的代码如下：
def img_from_fig(fig: matplotlib.figure.Figure) -> np.ndarray:
  """Returns a [H, W, 3] uint8 np image from fig.canvas.tostring_rgb()."""
  # Just enough margin in the figure to display xticks and yticks.
  fig.subplots_adjust(
      left=0.08, bottom=0.08, right=0.98, top=0.98, wspace=0.0, hspace=0.0
  )
  fig.canvas.draw()
  data = np.frombuffer(fig.canvas.tostring_argb(), dtype=np.uint8)
  r = data[1::4]  # 红色通道
  g = data[2::4]  # 绿色通道
  b = data[3::4]  # 蓝色通道
  img = np.stack([r, g, b], axis=-1).reshape(fig.canvas.get_width_height()[::-1] + (3,))
  plt.close(fig)
  return img

[the-big-bad-wolf]

You are using the latest version of the dataset right? I got the same error when trying to use version 1.3.0, but downgrading to the same version used in the tutorial and downgrading matplotlib fixed it.

[zuodong-zhong]

Yes, I am using the new version. To solve this issue, there are two methods: one is to downgrade the version, and the other is to use the ARGB format in the current version (i.e., modify waymax.utils.img_from_fig() by changing fig.canvas.tostring_rgb() to fig.canvas.tostring_argb()). However, directly modifying it leads to a new problem, specifically the error ValueError: cannot reshape array of size 1440000 into shape (600,600,3), due to data size mismatch.

This is because, after changing to the ARGB format, the image size becomes (600,600,4) (which gives 1440000 when multiplied), but in the subsequent code of waymax.utils.img_from_fig(), img = data.reshape(fig.canvas.get_width_height()[::-1] + (3,)) is used, where the target format is specified as (600,600,3).

Therefore, to solve this issue, there are two approaches:

1. Modify it to img = data.reshape(fig.canvas.get_width_height()[::-1] + (4,)), changing the target format to (600,600,4), but this will add transparency information to the generated image, making it appear pinkish.
2. After extracting the ARGB format, remove the transparency information to restore the original effect. You can modify it as follows:

data = np.frombuffer(fig.canvas.tostring_argb(), dtype=np.uint8)
r = data[1::4]  # Red channel
g = data[2::4]  # Green channel
b = data[3::4]  # Blue channel
img = np.stack([r, g, b], axis=-1).reshape(fig.canvas.get_width_height()[::-1] + (3,))

[zuodong-zhong]

Correct it, the error file is: waymax-main/waymax/visualization/utils.py中的img_from_fig(fig)函数

ValueError: cannot reshape array of size 1440000 into shape (600,600,3)

出错文件：waymax.utils.img_from_fig()

这是因为，waymax的utils.img_from_fig(fig)函数中，在处理图像时，先转成了RGB的一维数组形式，然后再重塑成图像的三维数组（h,w,r,g,b）格式，但我图像转换的函数库版本高，已经弃用了RGB的方法（所以我就换成了ARGB的格式，也就导致了上面的不匹配的错误），使用的是ARGB的格式（就是加了一个透明度），就无法适配指定的（h,w,r,g,b）格式了，所以我就把（h,w,r,g,b）格式改成了（h,w,r,g,b,a）格式，就画出来了，图片上应该是加上了透明度，修改后的代码如下： def img_from_fig(fig: matplotlib.figure.Figure) -> np.ndarray:   """Returns a [H, W, 3] uint8 np image from fig.canvas.tostring_rgb()."""   # Just enough margin in the figure to display xticks and yticks.   fig.subplots_adjust(       left=0.08, bottom=0.08, right=0.98, top=0.98, wspace=0.0, hspace=0.0   )   fig.canvas.draw()   data = np.frombuffer(fig.canvas.tostring_argb(), dtype=np.uint8)   r = data[1::4]  # 红色通道   g = data[2::4]  # 绿色通道   b = data[3::4]  # 蓝色通道   img = np.stack([r, g, b], axis=-1).reshape(fig.canvas.get_width_height()[::-1] + (3,))   plt.close(fig)   return img