Hi folks,
Delighted to share that we just landed efficient and performant TF segmentation models in 
Transformers. The model is called SegFormer, and it yields impressive performance while achieving efficiency. In particular, its performance on higher-resolution images (512x512, 640x640, etc.) is commendable. Below, I am sharing all the relevant links to help you get started:
- Model documentation: SegFormer
- Interactive space for playing with the model: SegFormer (ADE20k) in TensorFlow - a Hugging Face Space by chansung
- Fine-tune on your own segmentation dataset: https://github.com/deep-diver/segformer-tf-transformers
The demo and the notebooks were done in collaboration with @deep-diver. Happy to address any questions. Below is a short snippet of the results:
