LittleMeow

[here](https://github.com/AILab-CVC/YOLO-World/blob/da0fcb0ccf825e5fb9423651b11dfaac908f9249/yolo_world/models/detectors/yolo_world.py#L57) I-Pooling Attention is deprecated in v2

请字都不会说嘛...自己转个onnx再转coreml不就好了，后处理不支持的就跳过 自己写后处理

我不是官方人员，你可以自己把mmdeploy搞进这个工程里来，而不用easydeploy，mmdeploy可以直接转苹果。也可以跟着ncnn或者什么推理框架里的yolox yolov8，看cpp的后处理咋写的。

coco annotations.json要改成和你coco_class_texts.json保持一致。类别数越少讲道理会更好吧，用极限法，一个模型检测1个类和检测1万个类，肯定前者这个类效果更好 。。但具体到coco里指标能好多少 ，估计也不多

要删呀，category_id都变了 ，json文件重新生成一个就好了

从0开始

this [line](https://github.com/AILab-CVC/YOLO-World/blob/5ee2e01d52fd02aa75f731b145244ed204781d0f/deploy/export_onnx.py#L28)

1. I'm confused that in transform `YOLOv5RandomAffine`, `use_mask_refine` is [deprecated](https://github.com/onuralpszr/mmyolo/blob/4d97b3a06609dba94b8ec584be2f2029cfdb7519/mmyolo/datasets/transforms/transforms.py#L755) in your version of mmyolo. So, it should not influence the result ? 2. And, custom dataset usually dosen't have...

@wondervictor Thanks. For me, the problem is the decline of open-vocabulary ability after fine-tuning on custom dataset.

> @taofuyu, are the categories the same between the two models? yes, when exporting ONNX, I use the same categories as image_demo.py. You could verify this, maybe I mistake something.