Improve discoverability by pushing Co-Tracker to HF along with download stats

[NielsRogge]

Hi @nikitakaraevv and others,

Thanks for this nice work. I wrote a quick PoC to showcase that you can easily have integration on the 🤗 hub, which greatly improves the discoverability of your models.

This PR shows how you can automatically load the various CoTracker models using from_pretrained (and push them using push_to_hub), track download numbers for your models (similar to models in the Transformers library), and have nice model cards on a per-model basis. It leverages the PyTorchModelHubMixin class which allows to inherits these methods.

Usage is as follows (see also the notebook):

from cotracker.models.core.cotracker.cotracker import CoTracker2

model = CoTracker2.from_pretrained("nielsr/co-tracker-hf")

The corresponding model is here for now: https://huggingface.co/nielsr/co-tracker-hf. We could move all checkpoints to separate repos on the Meta organization if you're interested. Ideally, each checkpoint has its own model repository on the 🤗 hub, where a config.json and safetensors weights are hosted.

To improve discoverability, we could add tags like "object-tracking" to the model cards, so that people can find these models by either typing in the search bar/filtering on hf.co/models.

Kind regards,

Niels