def test_body_mesh_demo():
# H36M demo
config = 'configs/body/3d_mesh_sview_rgb_img/hmr' \
'/mixed/res50_mixed_224x224.py'
config = mmcv.Config.fromfile(config)
config.model.mesh_head.smpl_mean_params = \
'tests/data/smpl/smpl_mean_params.npz'
pose_model = None
with tempfile.TemporaryDirectory() as tmpdir:
config.model.smpl.smpl_path = tmpdir
config.model.smpl.joints_regressor = osp.join(
tmpdir, 'test_joint_regressor.npy')
# generate weight file for SMPL model.
generate_smpl_weight_file(tmpdir)
pose_model = init_pose_model(config, device='cpu')
assert pose_model is not None, 'Fail to build pose model'
image_name = 'tests/data/h36m/S1_Directions_1.54138969_000001.jpg'
det_result = {
'keypoints': np.zeros((17, 3)),
'bbox': [50, 50, 50, 50],
'image_name': image_name,
}
# make person bounding boxes
person_results = [det_result]
dataset = pose_model.cfg.data['test']['type']
# test a single image, with a list of bboxes
pose_results = inference_mesh_model(pose_model,
image_name,
person_results,
bbox_thr=None,
format='xywh',
dataset=dataset)
vis_3d_mesh_result(pose_model, pose_results, image_name)
def test_parametric_mesh_forward():
"""Test parametric mesh forward."""
tmpdir = tempfile.TemporaryDirectory()
# generate weight file for SMPL model.
generate_smpl_weight_file(tmpdir.name)
# Test ParametricMesh without discriminator
model_cfg = dict(
pretrained=None,
backbone=dict(type='ResNet', depth=50),
mesh_head=dict(
type='HMRMeshHead',
in_channels=2048,
smpl_mean_params='tests/data/smpl/smpl_mean_params.npz'),
disc=None,
smpl=dict(
type='SMPL',
smpl_path=tmpdir.name,
joints_regressor=osp.join(tmpdir.name,
'test_joint_regressor.npy')),
train_cfg=dict(disc_step=1),
test_cfg=dict(
flip_test=False,
post_process='default',
shift_heatmap=True,
modulate_kernel=11),
loss_mesh=dict(
type='MeshLoss',
joints_2d_loss_weight=1,
joints_3d_loss_weight=1,
vertex_loss_weight=1,
smpl_pose_loss_weight=1,
smpl_beta_loss_weight=1,
focal_length=5000,
img_res=256),
loss_gan=None)
detector = ParametricMesh(**model_cfg)
detector.init_weights()
optimizers_config = dict(generator=dict(type='Adam', lr=0.0001))
optims = build_optimizers(detector, optimizers_config)
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
# Test forward train
output = detector.train_step(mm_inputs, optims)
assert isinstance(output, dict)
# Test forward test
with torch.no_grad():
output = detector.val_step(data_batch=mm_inputs)
assert isinstance(output, dict)
imgs = mm_inputs.pop('img')
img_metas = mm_inputs.pop('img_metas')
output = detector.forward(imgs, img_metas=img_metas, return_loss=False)
assert isinstance(output, dict)
# Test ParametricMesh with discriminator
model_cfg['disc'] = dict()
model_cfg['loss_gan'] = dict(
type='GANLoss',
gan_type='lsgan',
real_label_val=1.0,
fake_label_val=0.0,
loss_weight=1)
optimizers_config['discriminator'] = dict(type='Adam', lr=0.0001)
detector = ParametricMesh(**model_cfg)
detector.init_weights()
optims = build_optimizers(detector, optimizers_config)
input_shape = (1, 3, 256, 256)
mm_inputs = _demo_mm_inputs(input_shape)
# Test forward train
output = detector.train_step(mm_inputs, optims)
assert isinstance(output, dict)
# Test forward test
with torch.no_grad():
output = detector.val_step(data_batch=mm_inputs)
assert isinstance(output, dict)
imgs = mm_inputs.pop('img')
img_metas = mm_inputs.pop('img_metas')
output = detector.forward(imgs, img_metas=img_metas, return_loss=False)
assert isinstance(output, dict)
_ = detector.forward_dummy(imgs)
tmpdir.cleanup()
def test_smpl():
"""Test smpl model."""
# build smpl model
smpl = None
with tempfile.TemporaryDirectory() as tmpdir:
# generate weight file for SMPL model.
generate_smpl_weight_file(tmpdir)
smpl_cfg = dict(smpl_path=tmpdir,
joints_regressor=osp.join(tmpdir,
'test_joint_regressor.npy'))
smpl = SMPL(**smpl_cfg)
assert smpl is not None, 'Fail to build SMPL model'
# test get face function
faces = smpl.get_faces()
assert isinstance(faces, np.ndarray)
betas = torch.zeros(3, 10)
body_pose = torch.zeros(3, 23 * 3)
global_orient = torch.zeros(3, 3)
transl = torch.zeros(3, 3)
gender = torch.LongTensor([-1, 0, 1])
# test forward with body_pose and global_orient in axis-angle format
smpl_out = smpl(betas=betas,
body_pose=body_pose,
global_orient=global_orient)
assert isinstance(smpl_out, dict)
assert smpl_out['vertices'].shape == torch.Size([3, 6890, 3])
assert smpl_out['joints'].shape == torch.Size([3, 24, 3])
# test forward with body_pose and global_orient in rotation matrix format
body_pose = torch.eye(3).repeat([3, 23, 1, 1])
global_orient = torch.eye(3).repeat([3, 1, 1, 1])
_ = smpl(betas=betas, body_pose=body_pose, global_orient=global_orient)
# test forward with translation
_ = smpl(betas=betas,
body_pose=body_pose,
global_orient=global_orient,
transl=transl)
# test forward with gender
_ = smpl(betas=betas,
body_pose=body_pose,
global_orient=global_orient,
transl=transl,
gender=gender)
# test forward when all samples in the same gender
gender = torch.LongTensor([0, 0, 0])
_ = smpl(betas=betas,
body_pose=body_pose,
global_orient=global_orient,
transl=transl,
gender=gender)
# test forward when batch size = 0
_ = smpl(betas=torch.zeros(0, 10),
body_pose=torch.zeros(0, 23 * 3),
global_orient=torch.zeros(0, 3))