This can be implemented similar to other pipelines (we already have one for zero-shot image classificaiton with CLIP, so it would be very similar to that one). For an example PR that added a pipeline, see #11598.
Motivation
Pipelines are great for abstracting away all the complexity for quick inference with a model.
As seen with @alaradirik this morning, this could also be leveraging the custom pipeline feature that was implemented last week, especially if this pipeline works with a very limited number of artchitectures.
Feature request
We currently have OWL-ViT in the library, which is capable of performing zero-shot object detection.
It would be great to have a pipeline for this task, with the following API:
This pipeline could default to the https://huggingface.co/google/owlvit-base-patch32 checkpoint. Also check out the demo notebook that showcases the model.
This can be implemented similar to other pipelines (we already have one for zero-shot image classificaiton with CLIP, so it would be very similar to that one). For an example PR that added a pipeline, see #11598.
Motivation
Pipelines are great for abstracting away all the complexity for quick inference with a model.
Your contribution
I can assist with this, together with @Narsil.
The text was updated successfully, but these errors were encountered: