Implement real-time semantic segmentation in tf 2.X

I want a working code for one of the models.:

1)STDC : Rethinking BiSeNet For Real-time Semantic Segmentation
2)FasterSeg: Searching for Faster Real-time Semantic Segmentation
3)BiSeNet V2:Rethinking BiSeNet For Real-time Semantic Segmentation

Model should be able to give me the output of city-scapes labels when image is passed.
There are lot of Github repositories available for all these models.
You can use any tensorflow 2.x implementation.
You don't have to train the model you can use available pretrained models.

Deliverables:
Source folder with inference code.
Drive link for downloading pre-trained weights you used for your testing.