search

Coldog2333 / pytoflow

The py version of toflow → https://github.com/anchen1011/toflow
MIT License
107 stars 31 forks source link

problems when using pytoflow #6

Closed mars11121 closed 4 years ago

mars11121 commented 4 years ago

when I use the code on Viemo90K, I have encountered a problem that the value range of processed images of the model is out of (0, 1), have you encountered this similar problem? and could you give me some help? thanks a lot!

Coldog2333 commented 4 years ago

Hello. I think you can print these images out to see whether they have something wrong. Sometimes the values maybe not so important, they can be rescaled when you output them as images.

mars11121 commented 4 years ago

okay, thanks a lot @Coldog2333; and I find that on line214, train.py, the check_point should be corrected to check_loss, which is used to replace original check_loss = 1.

Coldog2333 commented 4 years ago

Oh, thank you for your reminder! (It has been corrected.)

Coldog2333 commented 4 years ago

I remember that I checked all of the codes on my machine before pushing them. Maybe I should check it once again. (If I have time...)

mars11121 commented 4 years ago

fine, and your contribution is really useful for my work~

Coldog2333 commented 4 years ago

Glad to hear that. Thank you for your appreciation. :) If you have any other questions, just feel free to new an issue.

Coldog2333 commented 4 years ago

Hello, I check the codes of this repository today and figure out the reason for the problem you encountered. This is a bug or something like updated feature of matplotlib. If you use a matplotlib with version like 2.0.2, the problem will not occur.

reference: https://github.com/matterport/Mask_RCNN/issues/69