Error when trying to use sample code on CPU

[M3nxudo]

When trying to use the example code snippet in the Readme, i get an error, but it only happens when i force the device to cpu, with cuda acceleration i don't get the same error.

import cv2
import torch

from depth_anything_v2.dpt import DepthAnythingV2

model_configs = {
    'vits': {'encoder': 'vits', 'features': 64, 'out_channels': [48, 96, 192, 384]},
    'vitb': {'encoder': 'vitb', 'features': 128, 'out_channels': [96, 192, 384, 768]},
    'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
    'vitg': {'encoder': 'vitg', 'features': 384, 'out_channels': [1536, 1536, 1536, 1536]}
}

encoder = 'vitl'  # or 'vits', 'vitb', 'vitg'

model = DepthAnythingV2(**model_configs[encoder])
model.load_state_dict(torch.load(f'models/depth_anything_v2_{encoder}.pth', map_location='cpu'))
model.eval()

raw_img = cv2.imread('../folder/image.png')
depth = model.infer_image(raw_img)  # HxW raw depth map in numpy
cv2.imshow("Depth map",depth)
cv2.waitKey(0)

Only changed the directory for the models and for the image from the sample code, and i get the following error, seems to be a mismatch from trying to use cuda datatypes on cpu?

xFormers not available
xFormers not available
Traceback (most recent call last):
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\minimal.py", line 20, in <module>
    depth = model.infer_image(raw_img)  # HxW raw depth map in numpy
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\depth_anything_v2\dpt.py", line 190, in infer_image
    depth = self.forward(image)
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\depth_anything_v2\dpt.py", line 179, in forward
    features = self.pretrained.get_intermediate_layers(x, self.intermediate_layer_idx[self.encoder], return_class_token=True)
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\depth_anything_v2\dinov2.py", line 308, in get_intermediate_layers
    outputs = self._get_intermediate_layers_not_chunked(x, n)
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\depth_anything_v2\dinov2.py", line 272, in _get_intermediate_layers_not_chunked
    x = self.prepare_tokens_with_masks(x)
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\depth_anything_v2\dinov2.py", line 214, in prepare_tokens_with_masks
    x = self.patch_embed(x)
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\source\repos\Depth_Calibration\Depth-Anything-V2\depth_anything_v2\dinov2_layers\patch_embed.py", line 76, in forward
    x = self.proj(x)  # B C H W
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
    return forward_call(*args, **kwargs)
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\nn\modules\conv.py", line 460, in forward
    return self._conv_forward(input, self.weight, self.bias)
  File "C:\Users\Daniel\anaconda3\envs\GPU\lib\site-packages\torch\nn\modules\conv.py", line 456, in _conv_forward
    return F.conv2d(input, weight, bias, self.stride,
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

[M3nxudo]

I believe this is related to issue #15 , when modifying dpt.py with the suggested fix, the code works as expected now.

[LiheYoung]

Hi, we updated the readme: https://github.com/DepthAnything/Depth-Anything-V2#use-our-models

[jeezrick]

I don't think it's working.
Even if you set device to cpu in running script, in dpt.py file, input always gets transfer to cuda.

    
    def image2tensor(self, raw_image, input_size=518):        
        transform = Compose([
            Resize(
                width=input_size,
                height=input_size,
                resize_target=False,
                keep_aspect_ratio=True,
                ensure_multiple_of=14,
                resize_method='lower_bound',
                image_interpolation_method=cv2.INTER_CUBIC,
            ),
            NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
            PrepareForNet(),
        ])
        
        h, w = raw_image.shape[:2]
        
        image = cv2.cvtColor(raw_image, cv2.COLOR_BGR2RGB) / 255.0
        
        image = transform({'image': image})['image']
        image = torch.from_numpy(image).unsqueeze(0)
        
        DEVICE = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'
        image = image.to(DEVICE)
        
        return image, (h, w)

[y2w-oc]

I don't think it's working. Even if you set device to cpu in running script, in dpt.py file, input always gets transfer to cuda.

    
    def image2tensor(self, raw_image, input_size=518):        
        transform = Compose([
            Resize(
                width=input_size,
                height=input_size,
                resize_target=False,
                keep_aspect_ratio=True,
                ensure_multiple_of=14,
                resize_method='lower_bound',
                image_interpolation_method=cv2.INTER_CUBIC,
            ),
            NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
            PrepareForNet(),
        ])
        
        h, w = raw_image.shape[:2]
        
        image = cv2.cvtColor(raw_image, cv2.COLOR_BGR2RGB) / 255.0
        
        image = transform({'image': image})['image']
        image = torch.from_numpy(image).unsqueeze(0)
        
        DEVICE = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'
        image = image.to(DEVICE)
        
        return image, (h, w)

Hello! Have you solved this problem?