mohammaduzair9 / style-classification Goto Github PK

Hi I am using your project to discriminate between Signatures and Printed text in a document.
In order to train the data how many image samples of each class are required approximately?

Originally posted by @sahithi96 in #4 (comment)

I cloned your project and tried running it on documents but I'm not getting nice performance in that case.
The text detection part only detects words and it misses a lot text when you give it a inputs text-rich such a documents or books.
How could I change the model to fix this issue?

ValueError: Shape must be rank 1 but is rank 0 for 'bn_conv1/Reshape_4' (op: 'Reshape') with input shapes: [1,1,1,64], [].

Thank's for the awesome repo. I am thinking of retraining the classifier on my custom datasets, does it mean that I have to retrain frcnn too? I read the code and it seems that the feature of FPN is fed into the classifier to get the result. Is there a way to train the classifier only? Thank's a lot!

Hi,

It is a great work. I think the link to download the trained model is broken. If possible can you provide the updated link for the data or the trained model.

Hello, Could you please share the info about the dataset that you have used? Thank you

Thanks for the great work! I'm trying to get this to work with Keras 2.4 and encountering a few issues. The current one is as follows:

File "C:\Users\__\Anaconda3\envs\ocr\lib\site-packages\tensorflow\python\framework\tensor_util.py", line 463, in make_tensor_proto if shape is not None and np.prod(shape, dtype=np.int64) == 0: File "<__array_function__ internals>", line 6, in prod File "C:\Users\__\Anaconda3\envs\ocr\lib\site-packages\numpy\core\fromnumeric.py", line 3000, in prod keepdims=keepdims, initial=initial, where=where) File "C:\Users\__\Anaconda3\envs\ocr\lib\site-packages\numpy\core\fromnumeric.py", line 87, in _wrapreduction return ufunc.reduce(obj, axis, dtype, out, **passkwargs) TypeError: int() argument must be a string, a bytes-like object or a number, not 'NoneType'

This is happening due to a change in add_weight in FixedBatchNormalization.py where instead of passing shape, I pass (shape=shape)

Hello author, thank you very much for your research contribution, but the model download link is invalid, and can you give me a new model download address?

First of all thanks for this great repo,
I've tried to download your model but it seems the download link has expired, I'd like to know which Dataset are you using to train this model, Thanks.

Hi,its really a great work . can you please share the link for the same .

Dear author, the file of resnet50（resnet50_weights_tf_dim_ordering_tf_kernels.h5） does not exist under Style-Classification directory.

The model is taking more than 10s, when I am doing the test. Could you please help me with this?