KeyError: 'vit_h'

[LiYifan985]

F:\GS+\fenge\GaussianEditor\segment-anything-2\sam2\modeling\sam\transformer.py:23: UserWarning: You are using PyTorch 2.0.1+cu117 without Flash Attention v2 support. Consider upgrading to PyTorch 2.2+ for Flash Attention v2 (which could be faster).
OLD_GPU, USE_FLASH_ATTN, MATH_KERNEL_ON = get_sdpa_settings()
Reading camera 53/53
Traceback (most recent call last):
File "F:\GS+\fenge\GaussianEditor\webui.py", line 1569, in
webui = WebUI(args)
File "F:\GS+\fenge\GaussianEditor\webui.py", line 139, in init
self.text_segmentor = LangSAMTextSegmentor().to(get_device())
File "F:\GS+\fenge\GaussianEditor\threestudio\utils\sam.py", line 17, in init
self.model = LangSAM(sam_type)
File "F:\GS+\fenge\GaussianEditor\lang-segment-anything\lang_sam\lang_sam.py", line 12, in init
self.sam.build_model(sam_type, ckpt_path)
File "F:\GS+\fenge\GaussianEditor\lang-segment-anything\lang_sam\models\sam.py", line 44, in build_model
cfg = compose(config_name=SAM_MODELS[self.sam_type]["config"], overrides=[])
KeyError: 'vit_h'

[StardustandRain]

我也遇到了同样的问题 请问您解决了吗

[LiYifan985]

我也遇到了同样的问题 请问您解决了吗
你定位到源代码，有一个名称引用要更改，具体位置我忘了，在pycharm控制台运行根据报错位置找

[zhurui1995]

yeah, you can change "vit_h" to "sam2.1_hiera_large", it works for me.

threestudio/utils/sam.py

# line 15 
# def __init__(self, sam_type="vit_h"):
def __init__(self, sam_type="sam2.1_hiera_large"):
```