Qwen2-VL support detection

[WangRongsheng]

This will help you. In fact, qwen2-vl support object detection!

• https://github.com/QwenLM/Qwen2-VL/issues/9
• https://github.com/QwenLM/Qwen2-VL/issues/105
• https://github.com/QwenLM/Qwen2-VL/issues/190

[zhangfaen]

谢谢 Rongsheng提供的几个链接， 我快速看了一下里面的讨论， 发现 qwen-vl 和 qwen2-vl确实有考虑支持 object detection，但貌似qwen2-vl在object detection 性能上很差... 这个可能也是qwen2-vl发布的时候，没有刻意提 他们支持object detection的原因。 看起来 他们说 后续会提供一个方案。

谢谢 上面的几个链接， 很有帮助。

[WangRongsheng]

Maybe this pr had solved this issue.

https://github.com/huggingface/transformers/pull/33487/files

[zhangfaen]

Thank you Rongsheng. https://github.com/huggingface/transformers/pull/33487/files looks a reasonable fix. Hope it will be merged into main branch of transformers lib.

In this repo, I may cherry pick that PR first and have another try.

[WangRongsheng]

look forward to it.