LittleMeow

me too

> 到底放在哪个路径 YOLO-WORLD目录下建个文件夹pretrained_models。再翻墙把https://huggingface.co/openai/clip-vit-base-patch32/tree/main下载下来放进去即可

> > > 到底放在哪个路径 > > > > > > YOLO-WORLD目录下建个文件夹pretrained_models。再翻墙把[https://huggingface.co/openai/clip-vit-base-patch32/tree/main下载下来放进去即可](https://huggingface.co/openai/clip-vit-base-patch32/tree/main%E4%B8%8B%E8%BD%BD%E4%B8%8B%E6%9D%A5%E6%94%BE%E8%BF%9B%E5%8E%BB%E5%8D%B3%E5%8F%AF) > > 我已经下载下来了，但是还是不行，是哪里有问题  他config里写的是： text_model=dict( type='HuggingCLIPLanguageBackbone', model_name='pretrained_models/clip-vit-base-patch32-projection', frozen_modules=['all'])), 所以你下载的东西应该在pretrained_models/clip-vit-base-patch32-projection这个文件夹下面，你缺了个clip-vit-base-patch32-projection。 或者直接随便什么地方建个文件夹把东西放进去，再把这个model_name改了就行了

@tm924222 [__init__](https://github.com/AILab-CVC/YOLO-World/blob/4d90f458c1d0de310643b0ac2498f188c98c819c/yolo_world/datasets/__init__.py#L12)里写进去

@xiyangyang99 单步调试下吧，这种问题找起来很快的

> Yes, you are correct. 一个anchor预测的两个框的分数是不一样的，为啥可以通过复活相同分数框的做法来找回同组的预测呢？

So, can one anchor predict two different classes?

> @Purkialo has the implementation of a single bounding box and multiple labels implemented? can u share the code @taofuyu i am also interested in this did u see any...

Ok, I will have a try and update the result.

But, parameters of backbone, head, and neck are all frozen, and the only updated parameters 'embeddings' are not saved into hard disk (during inference, the pre-computed embedding file is still...