Molmo 2 (documentation: https://allenai.org/blog/molmo2 / models: https://huggingface.co/collections/allenai/molmo2 / demonstration: https://playground.allenai.org/?model=molmo2-8b) is a series of lightweight and multimodal models specialized in image and video analysis.

The models locate in space and count. One of them is dedicated to tracking objects on a video at up to 8 frames per second.

Open source, no accounts, free and unlimited.

Source for testing:

• Photo of Diego Girón: https://www.pexels.com/fr-fr/photo/un-troupeau-serein-d-oies-nageant-dans-les-eaux-bleues-33069798/
• Video by Daniel Torobekov: https://www.pexels.com/fr-fr/video/mer-personne-bleu-ocean-5743314/