def test_vipnas_forward():
# model settings
channel_cfg = dict(
num_output_channels=17,
dataset_joints=17,
dataset_channel=[
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
],
inference_channel=[
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
])
model_cfg = dict(type='TopDown',
pretrained=None,
backbone=dict(type='ViPNAS_ResNet', depth=50),
keypoint_head=dict(
type='ViPNASHeatmapSimpleHead',
in_channels=608,
out_channels=channel_cfg['num_output_channels'],
loss_keypoint=dict(type='JointsMSELoss',
use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
detector = TopDown(model_cfg['backbone'], None, model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'])
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(imgs,
target,
target_weight,
img_metas,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
def test_topdown_forward():
model_cfg = dict(type='TopDown',
pretrained=None,
backbone=dict(type='ResNet', depth=18),
keypoint_head=dict(
type='SimpleHead',
in_channels=512,
out_channels=17,
),
train_cfg=dict(),
test_cfg=dict(flip_test=False,
post_process=True,
shift_heatmap=True,
unbiased_decoding=False,
modulate_kernel=11),
loss_pose=dict(type='JointsMSELoss',
use_target_weight=False))
detector = TopDown(model_cfg['backbone'], model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'], model_cfg['loss_pose'])
detector.init_weights()
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(imgs,
target,
target_weight,
img_metas,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
detector = detector.cuda()
_ = detector.forward(imgs.cuda(),
img_metas=img_metas,
return_loss=False)
def test_topdown_forward():
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(type='ResNet', depth=18),
keypoint_head=dict(
type='TopdownHeatmapSimpleHead',
in_channels=512,
out_channels=17,
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
detector = TopDown(model_cfg['backbone'], None, model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'])
with pytest.raises(TypeError):
detector.init_weights(pretrained=dict())
detector.pretrained = model_cfg['pretrained']
detector.init_weights()
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(
imgs, target, target_weight, img_metas, return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
# flip test
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='HourglassNet',
num_stacks=1,
),
keypoint_head=dict(
type='TopdownHeatmapMultiStageHead',
in_channels=256,
out_channels=17,
num_stages=1,
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
loss_keypoint=dict(type='JointsMSELoss', use_target_weight=False)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
detector = TopDown(model_cfg['backbone'], None, model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'])
# Test forward train
losses = detector.forward(
imgs, target, target_weight, img_metas, return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='HourglassNet',
num_stacks=1,
),
keypoint_head=dict(
type='TopdownHeatmapMultiStageHead',
in_channels=256,
out_channels=17,
num_stages=1,
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
loss_keypoint=[
dict(
type='JointsMSELoss',
use_target_weight=True,
loss_weight=1.)
]),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=True,
modulate_kernel=11))
detector = TopDown(model_cfg['backbone'], None, model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'])
detector.init_weights()
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape, num_outputs=None)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(
imgs, target, target_weight, img_metas, return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='RSN',
unit_channels=256,
num_stages=1,
num_units=4,
num_blocks=[2, 2, 2, 2],
num_steps=4,
norm_cfg=dict(type='BN')),
keypoint_head=dict(
type='TopdownHeatmapMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=17,
num_stages=1,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN'),
loss_keypoint=[dict(type='JointsMSELoss', use_target_weight=True)]
* 3 + [dict(type='JointsOHKMMSELoss', use_target_weight=True)]),
train_cfg=dict(num_units=4),
test_cfg=dict(
flip_test=True,
post_process='default',
shift_heatmap=False,
unbiased_decoding=False,
modulate_kernel=5))
detector = TopDown(model_cfg['backbone'], None, model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'])
detector.init_weights()
input_shape = (1, 3, 256, 192)
mm_inputs = _demo_mm_inputs(input_shape, num_outputs=4)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(
imgs, target, target_weight, img_metas, return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
_ = detector.forward_dummy(imgs)
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(type='ResNet', depth=18),
neck=dict(type='GlobalAveragePooling'),
keypoint_head=dict(
type='DeepposeRegressionHead',
in_channels=512,
num_joints=17,
loss_keypoint=dict(type='SmoothL1Loss', use_target_weight=True)),
train_cfg=dict(),
test_cfg=dict(
flip_test=True,
regression_flip_shift=True,
))
detector = TopDown(model_cfg['backbone'], model_cfg['neck'],
model_cfg['keypoint_head'], model_cfg['train_cfg'],
model_cfg['test_cfg'], model_cfg['pretrained'])
with pytest.raises(TypeError):
detector.init_weights(pretrained=dict())
detector.pretrained = model_cfg['pretrained']
detector.init_weights()
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
imgs = mm_inputs.pop('imgs')
img_metas = mm_inputs.pop('img_metas')
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
def test_topdown_forward():
model_cfg = dict(type='TopDown',
pretrained=None,
backbone=dict(type='ResNet', depth=18),
keypoint_head=dict(
type='TopDownSimpleHead',
in_channels=512,
out_channels=17,
),
train_cfg=dict(),
test_cfg=dict(flip_test=False,
post_process=True,
shift_heatmap=True,
unbiased_decoding=False,
modulate_kernel=11),
loss_pose=dict(type='JointsMSELoss',
use_target_weight=True))
detector = TopDown(model_cfg['backbone'], model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'], model_cfg['loss_pose'])
detector.init_weights()
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(imgs,
target,
target_weight,
img_metas,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
# flip test
model_cfg = dict(type='TopDown',
pretrained=None,
backbone=dict(
type='HourglassNet',
num_stacks=1,
),
keypoint_head=dict(
type='TopDownMultiStageHead',
in_channels=256,
out_channels=17,
num_stages=1,
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
),
train_cfg=dict(),
test_cfg=dict(flip_test=True,
post_process=True,
shift_heatmap=True,
unbiased_decoding=False,
modulate_kernel=11),
loss_pose=dict(type='JointsMSELoss',
use_target_weight=False))
detector = TopDown(model_cfg['backbone'], model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'], model_cfg['loss_pose'])
# Test forward train
losses = detector.forward(imgs,
target,
target_weight,
img_metas,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(
type='HourglassNet',
num_stacks=1,
),
keypoint_head=dict(
type='TopDownMultiStageHead',
in_channels=256,
out_channels=17,
num_stages=1,
num_deconv_layers=0,
extra=dict(final_conv_kernel=1, ),
),
train_cfg=dict(loss_weights=([1])),
test_cfg=dict(flip_test=False,
post_process=True,
shift_heatmap=True,
unbiased_decoding=False,
modulate_kernel=11),
loss_pose=[dict(type='JointsMSELoss', use_target_weight=True)])
detector = TopDown(model_cfg['backbone'], model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'], model_cfg['loss_pose'])
detector.init_weights()
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape, num_outputs=1)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(imgs,
target,
target_weight,
img_metas,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
model_cfg = dict(
type='TopDown',
pretrained=None,
backbone=dict(type='RSN',
unit_channels=256,
num_stages=1,
num_units=4,
num_blocks=[2, 2, 2, 2],
num_steps=4,
norm_cfg=dict(type='BN')),
keypoint_head=dict(type='TopDownMSMUHead',
out_shape=(64, 48),
unit_channels=256,
out_channels=17,
num_stages=1,
num_units=4,
use_prm=False,
norm_cfg=dict(type='BN')),
train_cfg=dict(num_units=4),
test_cfg=dict(flip_test=True,
post_process=True,
shift_heatmap=False,
unbiased_decoding=False,
modulate_kernel=5),
loss_pose=[dict(type='JointsMSELoss', use_target_weight=True)] * 3 +
[dict(type='JointsOHKMMSELoss', use_target_weight=True)])
detector = TopDown(model_cfg['backbone'], model_cfg['keypoint_head'],
model_cfg['train_cfg'], model_cfg['test_cfg'],
model_cfg['pretrained'], model_cfg['loss_pose'])
detector.init_weights()
input_shape = (1, 3, 256, 192)
mm_inputs = _demo_mm_inputs(input_shape, num_outputs=4)
imgs = mm_inputs.pop('imgs')
target = mm_inputs.pop('target')
target_weight = mm_inputs.pop('target_weight')
img_metas = mm_inputs.pop('img_metas')
# Test forward train
losses = detector.forward(imgs,
target,
target_weight,
img_metas,
return_loss=True)
assert isinstance(losses, dict)
# Test forward test
with torch.no_grad():
_ = detector.forward(imgs, img_metas=img_metas, return_loss=False)
_ = detector.forward_dummy(imgs)