In your last commit, you removed the regularization loss:

3050a94

Why is that useful?
Thanks

The url to the weights file is incorrect, could release the correct url of your pretrained model? Thank you very much!

I tried to test the model on my own images but it doesn't work. I created the validation_data and images folders and used the inference notebook, but after I run the last line of code nothing happens.

Hi, and thanks for sharing your work.

Is it alright to adapt your code for a research paper (I'll add a reference to this github)?

If yes, you do need to add a license to your repo, though -> like
https://github.com/rishizek/tensorflow-deeplab-v3-plus/blob/master/LICENSE

Thanks!

Hello!
Thank you very much for this great implementation. Can you share the pretrained weights for this model ?

I get border artifacts like this:

Any ideas what could be causing it? I'm using my dataset (~1000 annotated images of the same box, out of the same video sequence). Note that if I use bigger batches and train for more epochs, the borders tend to get thinner but never disappear. They also appear randomly around the image.

Thank you fot your reply. I am a fresher. Now,I am training model with GPU.The batch_size is 12. Some results as follow:

As shown in the figure，the accuracy and dice_coef is too high,but the epoch is low . Can you tell me why?
Looking forward to your reply。

Can you provide the dataset link on which you trained your network so that I will try to apply it for our work on some lighter models like MobileNetV2

How to train with GPU because it takes too long to train with CPU.

How can i use it direct?

which dataset you used to train the network?
i faced the same issue recently. but, i dont know how to get the right human segmentation training dataset.
colud you give me some useful advices?. thx

hi，For person_segmentation i want to know which datasets do you use？
thanks