fastT5 VS OpenSeeFace

I've tried https://github.com/Ki6an/fastT5 but it works with CPU only.

I am working on a project where I fine-tuned a Pegasus model on the Reddit dataset. Now, I need to convert the fine-tuned model to ONNX for the deployment stage. I have followed this guide from Huggingface to convert to the ONNX model for unsupported architects. I got it done but the ONNX model can't generate text. Turned out that Pegasus is an encoder-decoder model and most guides are for either encoder-model (e.g. BERT) or decoder-model (e.g. GPT2). I found the only example of converting an encoder-decoder model to ONNX from here https://github.com/Ki6an/fastT5.

Microsoft Onnx Runtime T5 export tool / FastT5: to support caching, it exports 2 times the decoder part, one with cache, and one without (for the first generated token). So the memory footprint is doubled, which makes the solution difficult to use for these large transformer models.

Here is the fastT5 model code for reference code:https://github.com/Ki6an/fastT5/blob/master/fastT5/onnx_models.py

for more information on the project refer to the repository here.

I got something working modifying OpenSeeFace and it's an option and I might try to rewrite it in something compiled, but I'd like to look at the other options first.

Lastly, face tracking is either built in or uses a plugin device. You would also use OSC to manipulate blendshapes. I'd take a look at Opens Face.

Th best/only options I found were basically the old standby of Wine running closed source applications. Even then the end result was rather incomplete... because may of the best tracking options are simply closed source. (Note: Vseeface does offer an open source library)

There are only a few libraries that come to mind but take a bit of work to get started. MediaPipe Unity Plugin has eye tracking with a whole lot of types of tracking(head, hands, body). OpenSeeFace has models that do head and eye tracking. This repo uses Unity's neural net inference library, Barracuda, to run a MediaPipe iris landmark model (I haven't personally tested this library). Not sure how to translate eye landmarks to the position a player is looking at in a screen though. Hopefully this list of libraries gets you on the right path!

OpenSeeFace is open source. If you are using it with VSeeFace, you can just replace the Binary folder with your own build.

It uses OpenSeeFace for facetracking and engine patches/vrm code from the V-Sekai team.