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

[Frontend][TFLite] L2_POOL_2D operator #5452

Merged
merged 5 commits into from
Apr 29, 2020
Merged

[Frontend][TFLite] L2_POOL_2D operator #5452

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
bea7640
TFLITE fill and splitv ops
maheshambule Apr 20, 2020
8394633
Merge branch 'master' of https://github.com/apache/incubator-tvm into…
maheshambule Apr 27, 2020
9bac0b9
l2_pool_2d op changes in comment
maheshambule Apr 27, 2020
aa55c74
TFLite l2_pool_2d op added test case in main
maheshambule Apr 27, 2020
75caa3c
TFLite L2_POOL_2D added check for quantized input
maheshambule Apr 28, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions python/tvm/relay/frontend/tflite.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,7 @@ def __init__(self, model, subgraph, exp_tab):
'GREATER': self.convert_greater,
'HARD_SWISH': self.convert_hard_swish,
'L2_NORMALIZATION': self.convert_l2_normalization,
'L2_POOL_2D': self.convert_l2_pool2d,
'LESS_EQUAL': self.convert_less_equal,
'LESS': self.convert_less,
'LOCAL_RESPONSE_NORMALIZATION': self.convert_lrn,
Expand Down Expand Up @@ -331,6 +332,10 @@ def convert_max_pool2d(self, op):
"""Convert TFLite max pool2d"""
return self.convert_pool2d(op, "max")

def convert_l2_pool2d(self, op):
"""Convert TFLite l2 pool2d"""
return self.convert_pool2d(op, "l2")

def convert_reshape(self, op):
"""Convert TFLite reshape"""
try:
Expand Down Expand Up @@ -1674,6 +1679,12 @@ def convert_pool2d(self, op, pool_type):
assert self.has_same_qnn_params(input_tensor, output_tensor), \
"qnn.op.max_pool2d requires input and output qnn params to be same"
out = _op.nn.max_pool2d(in_expr, **params)
elif pool_type == "l2":
# l2_pool_2d is equivalent to sqrt(avg_pool(sqr(in_data)))
maheshambule marked this conversation as resolved.
Show resolved Hide resolved
exp_type = self.get_tensor_type_str(output_tensor.tensor.Type())
square_exp = _op.power(in_expr, relay.const(2, exp_type))
avg_pool_exp = _op.nn.avg_pool2d(square_exp, **params)
out = _op.sqrt(avg_pool_exp)
else:
raise tvm.error.OpNotImplemented(
'Operator {} is not supported for frontend TFLite.'.format(pool_type + ' pool'))
Expand Down
25 changes: 25 additions & 0 deletions tests/python/frontend/tflite/test_forward.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -487,6 +487,31 @@ def test_forward_pooling():
strides=[2, 1])


def _test_l2_pool2d(input_shape, ksize, strides, padding, data_format, fused_func_name=None):
x = np.arange(np.prod(input_shape), dtype=np.float32).reshape(input_shape) - 1

with tf.Graph().as_default():
in_data = tf.placeholder(
dtype=tf.float32, name="input", shape=input_shape)
out = tf.sqrt(tf.nn.avg_pool(
tf.square(in_data), ksize=ksize, strides=strides,
padding=padding, data_format=data_format))
out = with_fused_activation_function(out, fused_func_name)

compare_tflite_with_tvm(x, 'input', [in_data], [out])
Comment on lines +568 to +575
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Very quick drive-by review :
Don't we need a qnn test here ? I haven't managed to read up L2_POOL_2D in the tflite documentation yet.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

TFLite doesn't have support for quantized L2_POOL_2D yet.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks , possibly worth a belts and braces check to ensure we catch this in the frontend to ensure we think through qnn support rather than getting an unknown failure.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

added check.



def test_l2_pool2d():
maheshambule marked this conversation as resolved.
Show resolved Hide resolved
_test_l2_pool2d([1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], 'SAME', "NHWC", "RELU6")
_test_l2_pool2d([2, 9, 10, 2], [1, 1, 1, 1], [1, 1, 1, 1], 'SAME', "NHWC", "RELU6")
_test_l2_pool2d([2, 9, 10, 2], [1, 2, 1, 1], [1, 1, 1, 1], 'SAME', "NHWC")
_test_l2_pool2d([2, 9, 10, 2], [1, 2, 1, 1], [1, 1, 2, 1], 'SAME', "NHWC")
_test_l2_pool2d([1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], 'VALID', "NHWC", "RELU")
_test_l2_pool2d([2, 9, 10, 2], [1, 1, 1, 1], [1, 1, 1, 1], 'VALID', "NHWC")
_test_l2_pool2d([2, 9, 10, 2], [1, 2, 1, 1], [1, 1, 1, 1], 'VALID', "NHWC")
_test_l2_pool2d([2, 9, 10, 2], [1, 2, 1, 1], [1, 1, 2, 1], 'VALID', "NHWC", "RELU6")


#######################################################################
# Convolution
# -----------
Expand Down