Extraneous detection boxes on sample input using pre-trained weight file

[DamirWallener]

See attached images. It looks like the "actual" digits are being picked up correctly, but the results are getting clouded by a large number of '1.00' boxes. The two images reported:

18-boxes are detected
26-boxes are detected

That was with the threshold pushed all the way up to 0.99999. With the threshold at the original 0.3, the number of boxes is 64 and 87.

[gardethk]

Same problem here, i don't know what is causing this problem. I also tested it with another set of pictures with numbers, same result with the pre-trained weight file.

[penny4860]

The above phenomenon can not be reproduced in my execution environment. Please check keras version and tensorflow version.

[elfaizamine]

same issue here

[Amit12690]

I got the same issue , but when I installed the tensorflow , keras versions as mentioned in the readme , it worked perfectly. Looks like heavily dependent on the older version of tensorflow

[DamirWallener]

If there is that heavy a version dependency, that is highly problematic. It suggests the model isn't doing what we think it's doing.

…

On Mon, Jun 10, 2019, 8:38 AM amitbhatkal ***@***.***> wrote: I got the same issue , but when I installed the tensorflow , keras versions as mentioned in the readme , it worked perfectly. Looks like heavily dependent on the older version of tensorflow — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#27?email_source=notifications&email_token=ADAZWW7WCYFWAVWYDHRB63DPZZYPFA5CNFSM4HL5SAM2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXKHPDQ#issuecomment-500463502>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADAZWW4I3ZKG3RMXBT773FTPZZYPFANCNFSM4HL5SAMQ> .

[Amit12690]

With a bit of testing , I found that , it may only be dependent on the Keras version.
Gave extraneous boxes with keras 2.2.4
But worked perfectly after downgrading to keras 2.1.1 as indicated by the author irrespective of TensorFlow version (tried tf-1.12 )

[penny4860]

@Amit12690 thank you for testing

[manyaafonso]

I get the same problem even with the versions of tensorflow and keras being used by the author of the software. Any suggestion, @penny4860 ? I am using ubuntu, and i notice that you used windows. Could there be some dependency I am missing?

[esysss]

same issue here.