how to fine tune Qwen3.5 for object detection task (grounding 2D， LoRA)(如何使用LLama factory 进行Qwen3.5图像目标检测任务的LoRA微调？)

[WiedenWei]

Hi everyone,

Does anyone have a set of guidelines or best practices for this object detection fine-tuning task? I’m specifically looking for advice on optimal dataset formats and SFT (Supervised Fine-Tuning) configurations. Thanks in advance for your help!

各位大佬，

请帮帮忙吧，如何使用LLama factory 进行Qwen 3.5 模型进行图像目标检测任务？训练数据，脚本配置等有无参考的教程链接🙏🏻🙏🏻

[tshug]

同问。如何处理数据集？微调前，坐标是否需要归一化，如何归一化；微调后，推理是又需要对坐标如何处理？