A question about grayscale image

[HenryHZY]

Hi @jwyang I tested SoM+SAM with a script:

from PIL import Image
image = Image.open("gray.jpg")
output, mask = inference_sam_m2m_auto(model, image, text_size, label_mode, alpha, anno_mode)
image = Image.fromarray(output)
image.save("gray-som.jpg")

Running the script will throw some erros about image channel (e.g., image.shape[2]). It seems that SoM+SAM tool doesn't support grayscale image currently. Is there a mistake in my usage?

Does directly modifying some code to support grayscale maps affect the performance of SoM+SAM? For example:

image = image.convert('RGB')

[jwyang]

Hi, @HenryHZY, yes, adding the conversion line is a way of addressing the issue.

[HenryHZY]

Thanks a lot.