RuntimeError: expected scalar type float but found double about efficientsam

[hideinthesoul]

in efficient_sam.py, modify line 156 and 161:
def get_rescaled_pts(self, batched_points: torch.Tensor, input_h: int, input_w: int): return torch.stack( [ torch.where( batched_points[..., 0] >= 0, batched_points[..., 0] * self.image_encoder.img_size / input_w, **torch.tensor(-1.0, device=batched_points.device).float()**, ### modify -1 to this ), torch.where( batched_points[..., 1] >= 0, batched_points[..., 1] * self.image_encoder.img_size / input_h, **torch.tensor(-1.0, device=batched_points.device).float()**, ### modify -1 to this ), ], dim=-1, )

[yformer]

@TaoTXiXi, can you share which line causes that issue?

[TaoTXiXi]

Sure, this is where the error occurred.
Traceback (most recent call last):
File "G:/mycode/PythonCode/EfficientSAM-main/EfficientSAM_example.py", line 37, in
predicted_logits, predicted_iou = model(
File "D:\software\Miniconda3\envs\myProject\lib\site-packages\torch\nn\modules\module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "G:\mycode\PythonCode\EfficientSAM-main\efficient_sam\efficient_sam.py", line 207, in forward
return self.predict_masks(
File "G:\mycode\PythonCode\EfficientSAM-main\efficient_sam\efficient_sam.py", line 83, in predict_masks
rescaled_batched_points = self.get_rescaled_pts(batched_points, input_h, input_w)
File "G:\mycode\PythonCode\EfficientSAM-main\efficient_sam\efficient_sam.py", line 153, in get_rescaled_pts
torch.where(
RuntimeError: expected scalar type float but found double.

[yformer]

@TaoTXiXi, we checked EfficientSAM_example.py works well from our side. Can you create a new virtual env to rerun it?

[hjinnkim]

Hello! I have encountered exactly same issue with @TaoTXiXi . I'm using conda environment with torch==1.11 with cuda==11.3.

If you don't mind, can you share your virtual environment configurations @yformer ?

[hjinnkim]

Never mind, I solved the problem after upgrading my torch version from 1.11 to 1.12.1. I think older versions of torch have error with torch.where function.