def test_Combination():
dataset_params = {
'root': SHAPENET_ROOT,
'categories': ['can'],
'train': True,
'split': .8,
}
# images = shapenet.ShapeNet_Images(root=SHAPENET_ROOT, cache_dir=CACHE_DIR,
# categories=['bowl'], views=1, train=True, split=.8)
meshes = shapenet.ShapeNet_Meshes(**dataset_params)
voxels = shapenet.ShapeNet_Voxels(**dataset_params,
cache_dir=CACHE_DIR,
resolutions=[32])
sdf_points = shapenet.ShapeNet_SDF_Points(**dataset_params,
cache_dir=CACHE_DIR,
smoothing_iterations=3,
num_points=500,
occ=False,
sample_box=True)
points = shapenet.ShapeNet_Points(**dataset_params,
cache_dir=CACHE_DIR,
resolution=100,
smoothing_iterations=3,
num_points=500,
surface=False,
normals=True)
dataset = shapenet.ShapeNet_Combination([voxels, sdf_points, points])
for obj in dataset:
obj_data = obj['data']
assert set(obj['data']['sdf_points'].shape) == set([500, 3])
assert set(obj['data']['sdf_distances'].shape) == set([500])
assert set(obj['data']['32'].shape) == set([32, 32, 32])
assert set(obj['data']['points'].shape) == set([500, 3])
assert set(obj['data']['normals'].shape) == set([500, 3])
train_loader = DataLoader(dataset,
batch_size=2,
shuffle=True,
num_workers=8)
for batch in train_loader:
assert set(batch['data']['sdf_points'].shape) == set([2, 500, 3])
assert set(batch['data']['sdf_distances'].shape) == set([2, 500])
assert set(batch['data']['32'].shape) == set([2, 32, 32, 32])
assert set(batch['data']['points'].shape) == set([2, 500, 3])
assert set(batch['data']['normals'].shape) == set([2, 500, 3])
shutil.rmtree('tests/datasets/cache/sdf_points')
shutil.rmtree('tests/datasets/cache/points')
shutil.rmtree('tests/datasets/cache/voxels')
shutil.rmtree('tests/datasets/cache/surface_meshes')
split=.7,
num_points=5000)
images_set_valid = shapenet.ShapeNet_Images(
root='/media/archana/Local/Datasets/ShapeNetRendering',
categories=args.categories,
train=False,
split=.7,
views=1,
transform=preprocess)
meshes_set_valid = shapenet.ShapeNet_Meshes(
root='/media/archana/Local/Datasets/ShapeNetCore.v1.zip/ShapeNetCore.v1',
categories=args.categories,
train=False,
split=.7)
valid_set = shapenet.ShapeNet_Combination(
[points_set_valid, images_set_valid, meshes_set_valid])
dataloader_val = DataLoader(valid_set,
batch_size=args.batchsize,
shuffle=False,
collate_fn=collate_fn,
num_workers=0)
# Model
mesh = kal.rep.TriangleMesh.from_obj('386.obj', enable_adjacency=True)
mesh.cuda()
normalize_adj(mesh)
initial_verts = mesh.vertices.clone()
camera_fov_y = 49.13434207744484 * np.pi / 180.0
cam_proj = perspectiveprojectionnp(camera_fov_y, 1.0)
cam_proj = torch.FloatTensor(cam_proj).cuda()
# Data
mesh_set = shapenet.ShapeNet_Surface_Meshes(root=args.shapenet_root,
cache_dir=args.cache_dir,
categories=args.categories,
resolution=32,
train=False,
split=.7,
mode='Tri')
voxel_set = shapenet.ShapeNet_Voxels(root=args.shapenet_root,
cache_dir=args.cache_dir,
categories=args.categories,
train=False,
resolutions=[32],
split=.7)
valid_set = shapenet.ShapeNet_Combination([mesh_set, voxel_set])
encoder = MeshEncoder(30).to(args.device)
decoder = VoxelDecoder(30).to(args.device)
logdir = f'log/{args.expid}/AutoEncoder'
checkpoint = torch.load(os.path.join(logdir, 'best.ckpt'))
encoder.load_state_dict(checkpoint['encoder'])
decoder.load_state_dict(checkpoint['decoder'])
loss_epoch = 0.
num_batches = 0
num_items = 0
encoder.eval(), decoder.eval()
with torch.no_grad():
num_points=3000)
images_set = shapenet.ShapeNet_Images(root=args.shapenet_images_root,
categories=args.categories,
train=True,
split=.7,
views=23,
transform=preprocess)
if args.latent_loss:
mesh_set = shapenet.ShapeNet_Surface_Meshes(root=args.shapenet_root,
cache_dir=args.cache_dir,
categories=args.categories,
resolution=100,
train=True,
split=.7,
mode='Tri')
train_set = shapenet.ShapeNet_Combination(
[points_set, images_set, mesh_set])
dataloader_train = DataLoader(train_set,
batch_size=args.batch_size,
shuffle=True,
collate_fn=collate_fn,
num_workers=8)
else:
train_set = shapenet.ShapeNet_Combination([points_set, images_set])
dataloader_train = DataLoader(train_set,
batch_size=args.batch_size,
shuffle=True,
num_workers=8)
points_set_valid = shapenet.ShapeNet_Points(root=args.shapenet_root,
cache_dir=args.cache_dir,
categories=args.categories,
num_points=3000)
point_set = shapenet.ShapeNet_Points(
root='/media/archana/Local/Datasets/ShapeNetCore.v1.zip/ShapeNetCore.v1',
cache_dir='/media/archana/Local/Datasets/ShapeNetCore.v1.zip/cache/',
categories=args.categories,
train=True,
split=.7,
num_points=3000)
images_set = shapenet.ShapeNet_Images(
root='/media/archana/Local/Datasets/ShapeNetRendering',
categories=args.categories,
train=True,
split=.7,
views=23,
transform=preprocess)
train_set = shapenet.ShapeNet_Combination([sdf_set, images_set, point_set])
dataloader_train = DataLoader(train_set,
batch_size=args.batchsize,
shuffle=True,
num_workers=8)
sdf_set_valid = shapenet.ShapeNet_SDF_Points(
root='/media/archana/Local/Datasets/ShapeNetCore.v1.zip/ShapeNetCore.v1',
cache_dir='/media/archana/Local/Datasets/ShapeNetCore.v1.zip/cache/',
categories=args.categories,
train=False,
split=.2,
num_points=1000)
point_set_valid = shapenet.ShapeNet_Points(
root='/media/archana/Local/Datasets/ShapeNetCore.v1.zip/ShapeNetCore.v1',
