RuntimeError: Attempted to set the storage of a tensor on device "cpu" to a storage on different device "cuda:0". This is no longer allowed; the devices must match.

[liujiaoshou17]

RuntimeError: Attempted to set the storage of a tensor on device "cpu" to a storage on different device "cuda:0". This is no longer allowed; the devices must match.
使用预训练模型为imagenet时会报错，其他预训练模型就可以

[bayuezha]

在 hub.py里修改最后一行，删去, map_location=map_location

return torch.load(cached_file, weights_only=weights_only)

[LittleJJIU]

把serialization.py里这行代码删掉
tensor = torch.tensor([], dtype=storage.dtype).set_(
storage._storage, storage_offset, numel, stride)

改为：
确保存储（storage）和张量（tensor）都在 GPU 上
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

获取存储所在的设备
storage_device = storage._storage.device

如果存储已经在 GPU 上，直接使用其设备
if storage_device.type == "cuda":
tensor = torch.tensor([], dtype=storage.dtype, device=storage_device).set_(
storage.storage, storage_offset, numel, stride
)
else:
如果存储在 CPU 上，但你希望在 GPU 上训练，需要将存储移动到 GPU
tensor = torch.tensor([], dtype=storage.dtype, device=device).set(
storage._storage.to(device), storage_offset, numel, stride
)

[white-zz66]

手动修改patch源码
change_detection.pytorch-main\change_detection_pytorch\encoders_init_.py
将84行代码
encoder.load_state_dict(model_zoo.load_url(settings["url"], map_location=torch.device(DEVICE)))
修改成
encoder.load_state_dict(model_zoo.load_url(settings["url"]))

然后在脚本中添加
model = cdp.Unet(
encoder_name="resnet50",
encoder_weights="imagenet",
in_channels=3,
classes=2,
siam_encoder=True,
fusion_form='concat',
)
model.to(DEVICE) #add