Valid [0,1] data range check seems failed without eps considered for float datatype

[ziyannchen]

The error output

File "/root/anaconda3/envs/xxxx/lib/python3.8/site-packages/pyiqa/models/inference_model.py", line 79, in is_valid_input                                
    assert x.min() >= 0 and x.max() <= 1, f'Input must be normalized to [0, 1], but got min={x.min():.4f}, max={x.max():.4f}'                                
AssertionError: Input must be normalized to [0, 1], but got min=0.0000, max=1.0000

The source code

def is_valid_input(self, x):
        if x is not None:
            assert isinstance(x, torch.Tensor), 'Input must be a torch.Tensor'
            assert x.dim() == 4, 'Input must be 4D tensor (B, C, H, W)'
            assert x.shape[1] in [1, 3], 'Input must be RGB or gray image'
        
            if self.check_input_range:
                assert x.min() >= 0 and x.max() <= 1, f'Input must be normalized to [0, 1], but got min={x.min():.4f}, max={x.max():.4f}'

I checked the source code, it seems there is a problem when float datatype is given without eps considered.

[chaofengc]

Thank you so much for help. I will fix it in next update.

[chaofengc]

eps=1e-6 has been added to the range check code.

IQA-PyTorch/pyiqa/models/inference_model.py

Lines 82 to 83 in 61cf2fd

	if self.check_input_range:
	assert x.min() > -self.eps and x.max() < 1 + self.eps, f'Input must be normalized to [0, 1], but got min={x.min():.4f}, max={x.max():.4f}'