AttributeError: 'FigureCanvasAgg' object has no attribute 'tostring_rgb'

[Song-Pingfan]

Issue:

The following line in function show_factorization_on_image(img, explanations, colors = None, image_weight=0.5, concept_labels = None) gives AttributeError: 'FigureCanvasAgg' object has no attribute 'tostring_rgb'

data = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)

————————————————————————————
AttributeError Traceback (most recent call last)
in <cell line: 0>()
----> 1 visualization = show_factorization_on_image(rgb_img_float,
2 batch_explanations[0],
3 image_weight=0.3,
4 concept_labels=concept_label_strings)
5 result = np.hstack((img, visualization))

in show_factorization_on_image(img, explanations, colors, image_weight, concept_labels)
79 plt.axis('off')
80 fig.canvas.draw()
---> 81 data = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
82 # data = np.asarray(fig.canvas.buffer_rgba()) # data has shape (H, W, 4) (RGBA)
83 data = data[..., :3] # keep only RGB channels

AttributeError: 'FigureCanvasAgg' object has no attribute 'tostring_rgb'
————————————————————————————

Solution:

Try to use the canvas’s buffer to extract the image data.

Replace data = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8) by:

data = np.asarray(fig.canvas.buffer_rgba()) # data has shape (H, W, 4) (RGBA)
data = data[..., :3] # keep only RGB channels

Then, it works without any error.

[jacobgil]

Thank you! This was caused by maplotlib changes.
A PR will be greatly appreciated, otherwise I can test and add this in a few days.