Deployment of Custom Segmentation Models and Fine-tune Available Segmentation Models

[ali4864]

Is it possible to train custom image segmentation models and deploy on MediaPipe web? As I know we can use some .tflite models on MediaPipe, but it is a question that what are the constrains? What ops are valid to use with WebGL backend? What kind of architecture do it support?

And also is it possible to find-tune available models, like Selfie-Segmentation and Selfie-Segmentation-Multi-Class for web?