[kosmos-2] Is there a plan to release code for more tasks of kosmos-2?

[BIGBALLON]

trafficstars

Thanks for the amazing work [kosmos-2]. From the paper, kosmos-2 can do lots of tasks: (1) visual grounding, (2)-(3) grounded question answering, (4)-(6) multimodal referring via bounding boxes, and (7) grounded image captioning.

Is there any plan to release a demo for these tasks?

[pengzhiliang]

Thanks! Those all are in our release schedule. If you can't wait to experience it, please refer to here.

[BIGBALLON]

@pengzhiliang thanks a lot !!!

[pengzhiliang]

We have released evaluation code of some tasks in here.

[JierunChen]

We have released evaluation code of some tasks in here.

Thanks for your hard work. Could you continue to release the remaining evaluation code? It would be a valuable contribution to the field.

[donglixp]

We have released evaluation code of some tasks in here.

Thanks for your hard work. Could you continue to release the remaining evaluation code? It would be a valuable contribution to the field.

@JierunChen Which tasks you would like to evaluate first? So we could prioritize the release pipeline.

@pengzhiliang FYI.

[JierunChen]

@donglixp @pengzhiliang I do not have a strict preference and you may prioritize at your best convenience. The evaluation code is requested for the following tasks:

1. Referring expression generation
2. Image captioning
3. Visual question answering

Grateful for your contributions.