Jordan Pierce

@qiminchen, as an addition to a FCN for performing dense annotations, you may also be interested in [Multilevel Superpixel Segmentation](https://github.com/Shathe/ML-Superpixels), or [Fast Multilevel Superpixel Segmentation](https://github.com/JordanMakesMaps/Fast-Multilevel-Superpixel-Segmentation). These were both developed for...

What version of Keras do you have installed? Is it the same as the version required by the version of Segmentation Models you currently have installed?

> Hi all this is the correct requirements to make it working! working after multiple trial and errors. > !pip install keras==2.3.1 > !pip install tensorflow==2.1.0 > !pip install keras_applications==1.0.8...

Yes this is definitely possible. If you download imagenet weights from keras for resnet 50 and load them into the encoder portion of the unet model in the repo, it...

Like I said, I believe you need to create a custom function (though some might already exist on the internet, check stackoverflow), in which you go through each layer of...

Hi @tonyboston-au because the encode (i.e., resnet50) is within the Unet structure, keras treats it as a layer (thus ignoring the names of all of the resnet50 layer names). Can...

Also, see #36, they discuss the same thing but for a different reason.

My apologies! I see now that the UNet model doesn't treat the encoder as a layer (derp, not enough coffee) and understand the difficulties with the differences in naming. However,...

Search "negative loss" in the issues as there are already a few posts that describe potential reasons for why this is.

So normalizing (by dividing by 255) is really only helpful if your images are originally in the range of 0-255, and your model expects a range of 0-1. If you're...