def test_max_pool_bilinear_downsampling_cpu():
# Basic MappedTransposedConvolution layer
layer = MappedMaxPool(kernel_size=kernel_size).double()
# Run a forward and backward pass
output, forward_time, backward_time, gradcheck_res = utils.mapped_pool_test(
layer,
input=params.input_4x5().repeat(bs, 1, 1, 1),
sample_map=params.sample_map3(),
cuda=False)
# Manually computed correct result
correct_output = torch.tensor([[18.5, 16.5], [18.5, 18.5]]).double()
# Assert gradient check has passed
assert gradcheck_res
# Assert outputs match
testing.assert_allclose(output, correct_output)
def test_max_pool_integer_sampling_cpu():
# Basic MaxPool layer
layer = MappedMaxPool(kernel_size=kernel_size).double()
# Run a forward and backward pass
output, forward_time, backward_time, gradcheck_res = utils.mapped_pool_test(
layer,
input=params.input_4x5().repeat(bs, 1, 1, 1),
sample_map=params.sample_map0(),
cuda=False)
# Manually computed correct result
correct_output = torch.tensor([[11, 19, 14, 18, 13], [18, 17, 19, 18, 19],
[19, 12, 19, 10, 17], [18, 19, 10, 17,
17]]).double()
# Assert gradient check has passed
assert gradcheck_res
# Assert outputs match
testing.assert_allclose(output, correct_output)
def test_avg_pool_integer_sampling_cuda():
# Basic MappedTransposedConvolution layer
layer = MappedAvgPool(kernel_size=kernel_size).double().cuda()
# Run a forward and backward pass
output, forward_time, backward_time, gradcheck_res = utils.mapped_pool_test(
layer,
input=params.input_4x5().repeat(bs, 1, 1, 1),
sample_map=params.sample_map0(),
cuda=True)
# Manually computed correct result
correct_output = torch.tensor(
[[7.5, 6.25, 7.75, 9.75, 8.25], [12.25, 10., 10., 13.5, 10.75],
[11.5, 8.75, 11.75, 6.5, 8.25], [12.5, 9., 6.75, 10., 11.25]
], ).double().cuda()
# Assert gradient check has passed
assert gradcheck_res
# Assert outputs match
testing.assert_allclose(output, correct_output)
def test_max_pool_bilinear_interpolation_sampling_cuda():
# Basic MappedTransposedConvolution layer
layer = MappedMaxPool(kernel_size=kernel_size).double().cuda()
# Run a forward and backward pass
output, forward_time, backward_time, gradcheck_res = utils.mapped_pool_test(
layer,
input=params.input_4x5().repeat(bs, 1, 1, 1),
sample_map=params.sample_map1(),
cuda=True)
# Manually computed correct result
correct_output = torch.tensor([[14, 5, 8.25, 13, 16],
[9.25, 9, 9.25, 13,
13], [13, 15, 15, 13, 11],
[13, 13, 13, 13, 15]]).double().cuda()
# Assert gradient check has passed
assert gradcheck_res
# Assert outputs match
testing.assert_allclose(output, correct_output)