ValueError: operands could not be broadcast together with shapes (896,896,4) (3,) (896,896,4)

[JKYang01]

i am testing the model with different cases and something strange is, if I just feed it a screenshot .png picture of a table (sample attached) and I run the command python predict.py --model_weights='./tablenet_pretrained.ckpt' --image_path='./sample_table2.png'

it gives me value error operands could not be broadcast together with shapes (896,896,4) (3,) (896,896,4)

here is the full error log:

  File "/home/bluespinach/Documents/projects/tool_exp/OCR_tablenet/predict.py", line 148, in <module>
    predict()
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/click/core.py", line 1157, in __call__
    return self.main(*args, **kwargs)
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/click/core.py", line 1078, in main
    rv = self.invoke(ctx)
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/click/core.py", line 1434, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/click/core.py", line 783, in invoke
    return __callback(*args, **kwargs)
  File "/home/bluespinach/Documents/projects/tool_exp/OCR_tablenet/predict.py", line 144, in predict
    print(pred.predict(image))
  File "/home/bluespinach/Documents/projects/tool_exp/OCR_tablenet/predict.py", line 50, in predict
    processed_image = self.transforms(image=np.array(image))["image"]
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/albumentations/core/composition.py", line 182, in __call__
    data = t(force_apply=force_apply, **data)
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/albumentations/core/transforms_interface.py", line 89, in __call__
    return self.apply_with_params(params, **kwargs)
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/albumentations/core/transforms_interface.py", line 102, in apply_with_params
    res[key] = target_function(arg, **dict(params, **target_dependencies))
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/albumentations/augmentations/transforms.py", line 1496, in apply
    return F.normalize(image, self.mean, self.std, self.max_pixel_value)
  File "/home/bluespinach/miniconda3/envs/tablenet/lib/python3.9/site-packages/albumentations/augmentations/functional.py", line 141, in normalize
    img -= mean
ValueError: operands could not be broadcast together with shapes (896,896,4) (3,) (896,896,4) 

something strange is, if I input the .png file which is transferred from a pdf page, it works well. here is my testing sample, the pdf page

could you help to address why the smaller table screenshot picture doesn't work?

Thank you very much!

Best regards, JKyang01

[JKYang01]

some follow ups: I tried another way to crop the table. I transferred the pdf page into a .png file and save the table as a crop image of the pdf image. the output picture is like this: and I tried the detection again: python predict.py --model_weights='./tablenet_pretrained.ckpt' --image_path='./sample_table/test3_table_2.png'

it doesn't give me the error, but feed back with an empty list. []

That means the model cannot detect a table on the picture if the picture is the table (lol). I think probably the model is more use to see a table in side of a context (according to the training data).

Another thing I found is, if the pdf page have tables that connects to each other, the model will struggle to identify them as different tables, it will read those tables . and in data I am doing the processing, most of the pdf page have the tables that connects to each other.