Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[unit test] Assert failed in test_tinshift.py #1423

Closed
del-zhenwu opened this issue Oct 21, 2021 · 2 comments
Closed

[unit test] Assert failed in test_tinshift.py #1423

del-zhenwu opened this issue Oct 21, 2021 · 2 comments
Assignees
@grimoire

Comments

@del-zhenwu
Copy link
Collaborator

Checklist

  1. I have searched related issues but cannot get the expected help.
  2. I have read the FAQ documentation but cannot get the expected help.

Environment

Build full-mmcv version with the pytorch:pytorch image
image version: ubuntu_1804_py_37_cuda_101_cudnn_7_torch_160_dev
hardware: v100

10/21/2021 10:51:48 AM ____________________________ test_tinshift[dtype0] _____________________________
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM dtype = torch.float32
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM     @pytest.mark.skipif(
10/21/2021 10:51:48 AM         not torch.cuda.is_available(), reason='requires CUDA support')
10/21/2021 10:51:48 AM     @pytest.mark.parametrize('dtype', [torch.float, torch.double, torch.half])
10/21/2021 10:51:48 AM     def test_tinshift(dtype):
10/21/2021 10:51:48 AM >       _test_tinshift_allclose(dtype=dtype)
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM tests/test_ops/test_tin_shift.py:105:
10/21/2021 10:51:48 AM _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM dtype = torch.float32
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM     def _test_tinshift_allclose(dtype):
10/21/2021 10:51:48 AM         try:
10/21/2021 10:51:48 AM             from mmcv.ops import tin_shift
10/21/2021 10:51:48 AM         except ModuleNotFoundError:
10/21/2021 10:51:48 AM             pytest.skip('TinShift op is not successfully compiled')
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM         for shift, output, grad in zip(shifts, outputs, grads):
10/21/2021 10:51:48 AM             np_input = np.array(inputs)
10/21/2021 10:51:48 AM             np_shift = np.array(shift)
10/21/2021 10:51:48 AM             np_output = np.array(output)
10/21/2021 10:51:48 AM             np_grad = np.array(grad)
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM             x = torch.tensor(
10/21/2021 10:51:48 AM                 np_input, dtype=dtype, device='cuda', requires_grad=True)
10/21/2021 10:51:48 AM             shift = torch.tensor(np_shift, device='cuda').int()
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM             output = tin_shift(x, shift)
10/21/2021 10:51:48 AM             output.backward(torch.ones_like(output))
10/21/2021 10:51:48 AM >           assert np.allclose(
10/21/2021 10:51:48 AM                 output.data.type(torch.float).cpu().numpy(), np_output, 1e-3)
10/21/2021 10:51:48 AM E           AssertionError: assert False
10/21/2021 10:51:48 AM E            +  where False = <function allclose at 0x7f95d396def0>(array([[[[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n... ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]], dtype=float32), array([[[[ 0.4369, -3.7571],\n         [-1.1835, -1.6374],\n         [ 0.9534, -0.1321]],\n\n        [[-0.4658,  0.2162],\n...        [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]]), 0.001)
10/21/2021 10:51:48 AM E            +    where <function allclose at 0x7f95d396def0> = np.allclose
10/21/2021 10:51:48 AM E            +    and   array([[[[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n... ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]], dtype=float32) = <built-in method numpy of Tensor object at 0x7f95543ead20>()
10/21/2021 10:51:48 AM E            +      where <built-in method numpy of Tensor object at 0x7f95543ead20> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...     [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]]).numpy
10/21/2021 10:51:48 AM E            +        where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...     [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]]) = <built-in method cpu of Tensor object at 0x7f95543eac80>()
10/21/2021 10:51:48 AM E            +          where <built-in method cpu of Tensor object at 0x7f95543eac80> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0').cpu
10/21/2021 10:51:48 AM E            +            where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0') = <built-in method type of Tensor object at 0x7f95543eac80>(torch.float32)
10/21/2021 10:51:48 AM E            +              where <built-in method type of Tensor object at 0x7f95543eac80> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0').type
10/21/2021 10:51:48 AM E            +                where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0') = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...       [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0',\n       grad_fn=<TINShiftFunctionBackward>).data
10/21/2021 10:51:48 AM E            +              and   torch.float32 = torch.float
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM tests/test_ops/test_tin_shift.py:95: AssertionError
10/21/2021 10:51:48 AM ____________________________ test_tinshift[dtype1] _____________________________
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM dtype = torch.float64
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM     @pytest.mark.skipif(
10/21/2021 10:51:48 AM         not torch.cuda.is_available(), reason='requires CUDA support')
10/21/2021 10:51:48 AM     @pytest.mark.parametrize('dtype', [torch.float, torch.double, torch.half])
10/21/2021 10:51:48 AM     def test_tinshift(dtype):
10/21/2021 10:51:48 AM >       _test_tinshift_allclose(dtype=dtype)
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM tests/test_ops/test_tin_shift.py:105:
10/21/2021 10:51:48 AM _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM dtype = torch.float64
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM     def _test_tinshift_allclose(dtype):
10/21/2021 10:51:48 AM         try:
10/21/2021 10:51:48 AM             from mmcv.ops import tin_shift
10/21/2021 10:51:48 AM         except ModuleNotFoundError:
10/21/2021 10:51:48 AM             pytest.skip('TinShift op is not successfully compiled')
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM         for shift, output, grad in zip(shifts, outputs, grads):
10/21/2021 10:51:48 AM             np_input = np.array(inputs)
10/21/2021 10:51:48 AM             np_shift = np.array(shift)
10/21/2021 10:51:48 AM             np_output = np.array(output)
10/21/2021 10:51:48 AM             np_grad = np.array(grad)
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM             x = torch.tensor(
10/21/2021 10:51:48 AM                 np_input, dtype=dtype, device='cuda', requires_grad=True)
10/21/2021 10:51:48 AM             shift = torch.tensor(np_shift, device='cuda').int()
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM             output = tin_shift(x, shift)
10/21/2021 10:51:48 AM             output.backward(torch.ones_like(output))
10/21/2021 10:51:48 AM >           assert np.allclose(
10/21/2021 10:51:48 AM                 output.data.type(torch.float).cpu().numpy(), np_output, 1e-3)
10/21/2021 10:51:48 AM E           AssertionError: assert False
10/21/2021 10:51:48 AM E            +  where False = <function allclose at 0x7f95d396def0>(array([[[[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n... ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]], dtype=float32), array([[[[ 0.4369, -3.7571],\n         [-1.1835, -1.6374],\n         [ 0.9534, -0.1321]],\n\n        [[-0.4658,  0.2162],\n...        [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]]), 0.001)
10/21/2021 10:51:48 AM E            +    where <function allclose at 0x7f95d396def0> = np.allclose
10/21/2021 10:51:48 AM E            +    and   array([[[[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n... ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]], dtype=float32) = <built-in method numpy of Tensor object at 0x7f95543da0a0>()
10/21/2021 10:51:48 AM E            +      where <built-in method numpy of Tensor object at 0x7f95543da0a0> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...     [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]]).numpy
10/21/2021 10:51:48 AM E            +        where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...     [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]]) = <built-in method cpu of Tensor object at 0x7f95543da140>()
10/21/2021 10:51:48 AM E            +          where <built-in method cpu of Tensor object at 0x7f95543da140> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0').cpu
10/21/2021 10:51:48 AM E            +            where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0') = <built-in method type of Tensor object at 0x7f95544367d0>(torch.float32)
10/21/2021 10:51:48 AM E            +              where <built-in method type of Tensor object at 0x7f95544367d0> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000... 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0', dtype=torch.float64).type
10/21/2021 10:51:48 AM E            +                where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000... 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0', dtype=torch.float64) = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...000],\n          [ 0.0000,  0.0000]]]], device='cuda:0', dtype=torch.float64,\n       grad_fn=<TINShiftFunctionBackward>).data
10/21/2021 10:51:48 AM E            +              and   torch.float32 = torch.float
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM tests/test_ops/test_tin_shift.py:95: AssertionError
10/21/2021 10:51:48 AM ____________________________ test_tinshift[dtype2] _____________________________
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM dtype = torch.float16
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM     @pytest.mark.skipif(
10/21/2021 10:51:48 AM         not torch.cuda.is_available(), reason='requires CUDA support')
10/21/2021 10:51:48 AM     @pytest.mark.parametrize('dtype', [torch.float, torch.double, torch.half])
10/21/2021 10:51:48 AM     def test_tinshift(dtype):
10/21/2021 10:51:48 AM >       _test_tinshift_allclose(dtype=dtype)
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM tests/test_ops/test_tin_shift.py:105:
10/21/2021 10:51:48 AM _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM dtype = torch.float16
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM     def _test_tinshift_allclose(dtype):
10/21/2021 10:51:48 AM         try:
10/21/2021 10:51:48 AM             from mmcv.ops import tin_shift
10/21/2021 10:51:48 AM         except ModuleNotFoundError:
10/21/2021 10:51:48 AM             pytest.skip('TinShift op is not successfully compiled')
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM         for shift, output, grad in zip(shifts, outputs, grads):
10/21/2021 10:51:48 AM             np_input = np.array(inputs)
10/21/2021 10:51:48 AM             np_shift = np.array(shift)
10/21/2021 10:51:48 AM             np_output = np.array(output)
10/21/2021 10:51:48 AM             np_grad = np.array(grad)
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM             x = torch.tensor(
10/21/2021 10:51:48 AM                 np_input, dtype=dtype, device='cuda', requires_grad=True)
10/21/2021 10:51:48 AM             shift = torch.tensor(np_shift, device='cuda').int()
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM             output = tin_shift(x, shift)
10/21/2021 10:51:48 AM             output.backward(torch.ones_like(output))
10/21/2021 10:51:48 AM >           assert np.allclose(
10/21/2021 10:51:48 AM                 output.data.type(torch.float).cpu().numpy(), np_output, 1e-3)
10/21/2021 10:51:48 AM E           AssertionError: assert False
10/21/2021 10:51:48 AM E            +  where False = <function allclose at 0x7f95d396def0>(array([[[[ 0.        ,  0.        ],\n         [ 0.        ,  0.        ],\n         [ 0.        ,  0.        ]],\n\n     ...[ 0.        ,  0.        ],\n         [ 0.        ,  0.        ],\n         [ 0.        ,  0.        ]]]], dtype=float32), array([[[[ 0.4369, -3.7571],\n         [-1.1835, -1.6374],\n         [ 0.9534, -0.1321]],\n\n        [[-0.4658,  0.2162],\n...        [ 0.    ,  0.    ]],\n\n        [[ 0.    ,  0.    ],\n         [ 0.    ,  0.    ],\n         [ 0.    ,  0.    ]]]]), 0.001)
10/21/2021 10:51:48 AM E            +    where <function allclose at 0x7f95d396def0> = np.allclose
10/21/2021 10:51:48 AM E            +    and   array([[[[ 0.        ,  0.        ],\n         [ 0.        ,  0.        ],\n         [ 0.        ,  0.        ]],\n\n     ...[ 0.        ,  0.        ],\n         [ 0.        ,  0.        ],\n         [ 0.        ,  0.        ]]]], dtype=float32) = <built-in method numpy of Tensor object at 0x7f955432c1e0>()
10/21/2021 10:51:48 AM E            +      where <built-in method numpy of Tensor object at 0x7f955432c1e0> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...     [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]]).numpy
10/21/2021 10:51:48 AM E            +        where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...     [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]]) = <built-in method cpu of Tensor object at 0x7f955432c820>()
10/21/2021 10:51:48 AM E            +          where <built-in method cpu of Tensor object at 0x7f955432c820> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0').cpu
10/21/2021 10:51:48 AM E            +            where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000....0000]],\n\n         [[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0') = <built-in method type of Tensor object at 0x7f955432c370>(torch.float32)
10/21/2021 10:51:48 AM E            +              where <built-in method type of Tensor object at 0x7f955432c370> = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000... 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0', dtype=torch.float16).type
10/21/2021 10:51:48 AM E            +                where tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000... 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]]]], device='cuda:0', dtype=torch.float16) = tensor([[[[ 0.0000,  0.0000],\n          [ 0.0000,  0.0000],\n          [ 0.0000,  0.0000]],\n\n         [[ 0.0000,  0.000...000],\n          [ 0.0000,  0.0000]]]], device='cuda:0', dtype=torch.float16,\n       grad_fn=<TINShiftFunctionBackward>).data
10/21/2021 10:51:48 AM E            +              and   torch.float32 = torch.float
10/21/2021 10:51:48 AM 
10/21/2021 10:51:48 AM tests/test_ops/test_tin_shift.py:95: AssertionError
@grimoire
Copy link
Member

Hi, thanks for the report.
There are some problems with the current test data. We will update soon.

@zhouzaida
Copy link
Collaborator

closed by #1426

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@grimoire grimoire

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@grimoire @del-zhenwu @zhouzaida