|
|
|
import numpy as np |
|
import pytest |
|
import torch |
|
|
|
from mmpose.models.necks import GlobalAveragePooling |
|
|
|
|
|
def test_gap(): |
|
"""Test GlobalAveragePooling neck.""" |
|
gap = GlobalAveragePooling() |
|
|
|
with pytest.raises(TypeError): |
|
gap(1) |
|
|
|
x0_shape = (32, 1024, 4, 4) |
|
x1_shape = (32, 2048, 2, 2) |
|
x0 = _demo_inputs(x0_shape) |
|
x1 = _demo_inputs(x1_shape) |
|
|
|
y = gap(x0) |
|
assert y.shape == torch.Size([32, 1024]) |
|
|
|
y = gap([x0, x1]) |
|
assert y[0].shape == torch.Size([32, 1024]) |
|
assert y[1].shape == torch.Size([32, 2048]) |
|
|
|
y = gap((x0, x1)) |
|
assert y[0].shape == torch.Size([32, 1024]) |
|
assert y[1].shape == torch.Size([32, 2048]) |
|
|
|
|
|
def _demo_inputs(input_shape=(1, 3, 64, 64)): |
|
"""Create a superset of inputs needed to run backbone. |
|
|
|
Args: |
|
input_shape (tuple): input batch dimensions. |
|
Default: (1, 3, 64, 64). |
|
""" |
|
imgs = np.random.random(input_shape) |
|
imgs = torch.FloatTensor(imgs) |
|
|
|
return imgs |
|
|
