some trivial problems

[GNAYUOHZ]

❓ Questions and Help

Thank you for your excelent work.
I used the pretrained mask_rcnn_X_101_32x8d_FPN_1x model to infer this picture --->>>>"http://farm3.staticflickr.com/2469/3915380994_2e611b1779_z.jpg" . However the result is worse than the one shown in readme. Are there some parameters I need to change except confidence_threshold?
On the other hand, the font and mask displayed are ugly. Do you plan to optimize the display effect?
On the other other hand,I think the code structure is not very clear. Will there be any changes later?

[fmassa]

Hi,

Thanks for the feedback.

The only thing I changed for the demo was the input size, and I added it to be 800 (which is the default value for training). Apart from that, the rest should be the same.

About the display being ugly, I'd definitely accept PRs improving the font and the size!

Code structure not being clear: yes, I plan to improve it further. It would be very helpful if you could point out the parts that are not clear to you.

Thanks!

[GNAYUOHZ]

Thank you for response.
I think it's a little redundant some places, for example mkdir doesn't have to write a py file to check for existence, but can just use os. makedirs (***, exist_ok = True) instead？
I like to simplify the code.I'm just a little obsessive compulsive disorder. hhhh
just some of my humble opinions.

[fmassa]

Hi,

About the mkdirs having the exist_ok argument, this is unfortunately a python 3 feature, and doesn't exist on Python 2.
We are supporting Python2 as well, so we needed to use a custom solution for that particular case. But maybe a comment on the function might be good to do, explaining that we need to maintain py2 compact.

Please let us know if there are other things that you'd think could be improved.