def full_eval(opt):
indir_root = opt.indir
outdir_root = os.path.join(opt.outdir, opt.models+os.path.splitext(opt.modelpostfix)[0])
datasets = opt.dataset
if not isinstance(datasets, list):
datasets = [datasets]
for dataset in datasets:
print(f'Evaluating on dataset {dataset}')
opt.indir = os.path.join(indir_root, os.path.dirname(dataset))
opt.outdir = os.path.join(outdir_root, os.path.dirname(dataset))
opt.dataset = os.path.basename(dataset)
# evaluate
if os.path.exists(os.path.join(opt.indir, '05_query_dist')):
opt.reconstruction = False
points_to_surf_eval.eval_meshnet(opt)
res_dir_eval = os.path.join(opt.outdir, 'eval')
evaluation.eval_predictions(
os.path.join(res_dir_eval, 'eval'),
os.path.join(opt.indir, '05_query_dist'),
os.path.join(res_dir_eval, 'rme_comp_res.csv'),
unsigned=False)
# reconstruct
start = time.time()
opt.reconstruction = True
points_to_surf_eval.eval_meshnet(opt)
res_dir_rec = os.path.join(opt.outdir, 'rec')
end = time.time()
print('Inference of SDF took: {}'.format(end - start))
start = time.time()
imp_surf_dist_ms_dir = os.path.join(res_dir_rec, 'dist_ms')
query_pts_ms_dir = os.path.join(res_dir_rec, 'query_pts_ms')
vol_out_dir = os.path.join(res_dir_rec, 'vol')
mesh_out_dir = os.path.join(res_dir_rec, 'mesh')
sdf.implicit_surface_to_mesh_directory(
imp_surf_dist_ms_dir, query_pts_ms_dir,
vol_out_dir, mesh_out_dir,
opt.query_grid_resolution,
opt.sigma,
opt.certainty_threshold,
opt.workers)
end = time.time()
print('Sign propagation took: {}'.format(end - start))
new_meshes_dir_abs = os.path.join(res_dir_rec, 'mesh')
ref_meshes_dir_abs = os.path.join(opt.indir, '03_meshes')
csv_file = os.path.join(res_dir_rec, 'hausdorff_dist_pred_rec.csv')
evaluation.mesh_comparison(
new_meshes_dir_abs=new_meshes_dir_abs,
ref_meshes_dir_abs=ref_meshes_dir_abs,
num_processes=opt.workers,
report_name=csv_file,
samples_per_model=10000,
dataset_file_abs=os.path.join(opt.indir, opt.dataset))
# reconstructed_meshes_raw = '/home/perler/repos/AtlasNet/log/atlasnet_ae_imp_surf_14_25_squares/' + dataset + '/'
# reconstructed_meshes = '/home/perler/repos/AtlasNet/log/atlasnet_ae_imp_surf_14_25_squares/' + dataset + '/ply/'
# report_file = '/home/perler/repos/AtlasNet/log/atlasnet_ae_imp_surf_14_25_squares/' + dataset + '.csv'
#
# # for AtlasNet
# revert_atlasnet_transform(
# in_dir_abs=reconstructed_meshes_raw,
# out_dir_abs=reconstructed_meshes,
# ref_meshes_dir_abs=original_meshes,
# #ref_meshes_dir_abs=original_pc, # for real-world, no mesh exists
# num_processes=num_processes)
#
# utils_eval.mesh_comparison(
# new_meshes_dir_abs=reconstructed_meshes, ref_meshes_dir_abs=original_meshes,
# num_processes=num_processes, report_name=report_file,
# samples_per_model=10000, dataset_file_abs=val_set_file_abs)
# DeepSDF train imp_surf_14, val imp_surf_14
original_meshes = '/data/datasets/own/' + dataset + '/03_meshes'
val_set_file_abs = '/data/datasets/own/' + dataset + '/testset.txt'
reconstructed_meshes = '/home/perler/repos/DeepSDF/examples/' + dataset + '/Reconstructions/1000/Meshes/' + dataset + '/03_meshes/'
report_file = '/home/perler/repos/DeepSDF/examples/' + dataset + '/Reconstructions/1000/Meshes/' + dataset + '/imp_surf_14_comp.csv'
print('### chamfer distance: {}'.format(dataset))
evaluation.mesh_comparison(new_meshes_dir_abs=reconstructed_meshes,
ref_meshes_dir_abs=original_meshes,
num_processes=num_processes,
report_name=report_file,
samples_per_model=10000,
dataset_file_abs=val_set_file_abs)
def main(dataset_name: str):
# meshlabserver = "C:\\Program Files\\VCG\\MeshLab\\meshlabserver.exe"
meshlabserver = '/home/perler/repos/meshlab/src/distrib/meshlabserver'
num_processes = 12
# num_processes = 1
# base_dir = '../../datasets'
base_dir = '/data/datasets/own/'
#dataset_dir = 'implicit_surf_8'
#dataset_dir = 'implicit_surf_real_world'
dataset_dir = dataset_name
config_file = os.path.join(base_dir, dataset_dir, 'settings.ini')
config = configparser.ConfigParser()
read_config(config, config_file)
print('Processing dataset: ' + config_file)
do_clean = False
filter_broken_inputs = True
dirs_to_clean = \
['00_base_meshes',
'01_base_meshes_ply',
'02_meshes_cleaned',
'03_meshes',
'04_pts', '04_blensor_py', # , '04_pcd' '04_pts_noisefree',
'05_patch_dists', '05_patch_ids', '05_query_dist', '05_query_pts',
'05_patch_ids_grid', '05_query_pts_grid', '05_query_dist_grid',
'06_poisson_rec', '06_mc_gt_recon', '06_poisson_rec_gt_normals',
'06_normals', '06_normals/pts', '06_dist_from_p_normals']
dirs_to_clean_remainders = \
['00_base_meshes',
'01_base_meshes_ply',
'02_meshes_cleaned',
'03_meshes',
'04_pts', '04_blensor_py', # '04_pcd', '04_pts_noisefree',
'05_patch_dists', '05_patch_ids', '05_query_dist', '05_query_pts',
'05_patch_ids_grid', '05_query_pts_grid', '05_query_dist_grid',
'06_poisson_rec', '06_mc_gt_recon', '06_poisson_rec_gt_normals',
'06_normals', '06_normals/pts', '06_dist_from_p_normals']
# clean old dataset
if do_clean:
for dir in dirs_to_clean:
shutil.rmtree(os.path.join(base_dir, dataset_dir, dir), True)
if filter_broken_inputs: # the user might have removed unwanted input meshes after some processing
clean_up_broken_inputs(base_dir=base_dir,
dataset_dir=dataset_dir,
final_out_dir='00_base_meshes',
final_out_extension=None,
clean_up_dirs=dirs_to_clean_remainders,
broken_dir='broken')
start = time.time()
print('### reconstruct poisson with pcpnet normals')
dirs = (
os.path.join(base_dir, dataset_dir, '04_pts_vis'),
os.path.join(base_dir, dataset_dir, '06_normals_pcpnet'),
)
endings_per_dir = (
'.xyz',
'.normals',
)
file_utils.concat_txt_dirs(ref_dir=os.path.join(base_dir, dataset_dir,
'06_normals_pcpnet'),
ref_ending='.normals',
dirs=dirs,
endings_per_dir=endings_per_dir,
out_dir=os.path.join(base_dir, dataset_dir,
'07_pts_normals_pcpnet'),
out_ending='.xyz')
print('### poisson reconstruction from pcpnet normals')
apply_meshlab_filter(base_dir=base_dir,
dataset_dir=dataset_dir,
pts_dir='07_pts_normals_pcpnet',
recon_mesh_dir='06_poisson_rec_pcpnet_normals',
num_processes=num_processes,
filter_file='poisson.mlx',
meshlabserver_bin=meshlabserver)
end = time.time()
print('SPSR with PCPNet normals took: {}'.format(end - start))
print(
'### normal estimation and poisson reconstruction pcpnet - hausdorff distance'
)
new_meshes_dir_abs = os.path.join(base_dir, dataset_dir,
'06_poisson_rec_pcpnet_normals')
ref_meshes_dir_abs = os.path.join(base_dir, dataset_dir, '03_meshes')
csv_file = os.path.join(base_dir, dataset_dir,
'comp_poisson_rec_pcpnet_normals.csv')
val_set_file_abs = os.path.join(base_dir, dataset_dir, 'valset.txt')
evaluation.mesh_comparison(new_meshes_dir_abs=new_meshes_dir_abs,
ref_meshes_dir_abs=ref_meshes_dir_abs,
num_processes=num_processes,
report_name=csv_file,
samples_per_model=10000,
dataset_file_abs=val_set_file_abs)
print('### get ground truth normals for point cloud')
utils.get_pts_normals(base_dir=base_dir,
dataset_dir=dataset_dir,
dir_in_pointcloud='04_pts',
dir_in_meshes='03_meshes',
dir_out_normals='06_normals',
samples_per_model=100000,
num_processes=num_processes)
print('### poisson reconstruction from gt normals')
apply_meshlab_filter(base_dir=base_dir,
dataset_dir=dataset_dir,
pts_dir='06_normals/pts',
recon_mesh_dir='06_poisson_rec_gt_normals',
num_processes=num_processes,
filter_file='../../poisson.mlx',
meshlabserver_bin=meshlabserver)
print(
'### normal estimation and poisson reconstruction gt normals - hausdorff distance'
)
new_meshes_dir_abs = os.path.join(base_dir, dataset_dir,
'06_poisson_rec_gt_normals')
ref_meshes_dir_abs = os.path.join(base_dir, dataset_dir, '03_meshes')
csv_file = os.path.join(base_dir, dataset_dir,
'comp_poisson_rec_gt_normals.csv')
val_set_file_abs = os.path.join(base_dir, dataset_dir, 'valset.txt')
evaluation.mesh_comparison(new_meshes_dir_abs=new_meshes_dir_abs,
ref_meshes_dir_abs=ref_meshes_dir_abs,
num_processes=num_processes,
report_name=csv_file,
samples_per_model=10000,
dataset_file_abs=val_set_file_abs)
'--epsilon',
str(rec_epsilon),
]
recon_opt = points_to_surf_eval.parse_arguments(recon_params)
points_to_surf_eval.points_to_surf_eval(recon_opt)
# reconstruct meshes from predicted SDFs
imp_surf_dist_ms_dir = os.path.join(res_dir_rec, 'dist_ms')
query_pts_ms_dir = os.path.join(res_dir_rec, 'query_pts_ms')
vol_out_dir = os.path.join(res_dir_rec, 'vol')
mesh_out_dir = os.path.join(res_dir_rec, 'mesh')
sdf.implicit_surface_to_mesh_directory(imp_surf_dist_ms_dir,
query_pts_ms_dir, vol_out_dir,
mesh_out_dir, grid_resolution,
sigma, certainty_threshold,
workers)
# get Hausdorff distance for reconstructed meshes
new_meshes_dir_abs = os.path.join(res_dir_rec, 'mesh')
ref_meshes_dir_abs = os.path.join(in_dir_test, '03_meshes')
csv_file = os.path.join(res_dir_rec, 'hausdorff_dist_pred_rec.csv')
evaluation.mesh_comparison(new_meshes_dir_abs=new_meshes_dir_abs,
ref_meshes_dir_abs=ref_meshes_dir_abs,
num_processes=workers,
report_name=csv_file,
samples_per_model=10000,
dataset_file_abs=os.path.join(
in_dir_test, test_set))
print('Points2Surf is finished!')
def main(dataset_name: str):
# meshlabserver = "C:\\Program Files\\VCG\\MeshLab\\meshlabserver.exe"
meshlabserver = '~/repos/meshlab/src/distrib/meshlabserver'
num_processes = 12
# num_processes = 1
base_dir = 'datasets'
dataset_dir = dataset_name
print('Processing dataset: ' + dataset_name)
filter_broken_inputs = True
dirs_to_clean = \
['00_base_meshes',
'01_base_meshes_ply',
'02_meshes_cleaned',
'03_meshes',
'04_pts', '04_blensor_py',
'05_patch_dists', '05_patch_ids', '05_query_dist', '05_query_pts',
'05_patch_ids_grid', '05_query_pts_grid', '05_query_dist_grid',
'06_poisson_rec', '06_mc_gt_recon', '06_poisson_rec_gt_normals',
'06_normals', '06_normals/pts', '06_dist_from_p_normals']
if filter_broken_inputs: # the user might have removed unwanted input meshes after some processing
clean_up_broken_inputs(base_dir=base_dir, dataset_dir=dataset_dir,
final_out_dir='00_base_meshes', final_out_extension=None,
clean_up_dirs=dirs_to_clean, broken_dir='broken')
start = time.time()
print('### reconstruct poisson with pcpnet normals')
dirs = (os.path.join(base_dir, dataset_dir, '04_pts_vis'),
os.path.join(base_dir, dataset_dir, '06_normals_pcpnet'),)
endings_per_dir = ('.xyz', '.normals', )
file_utils.concat_txt_dirs(
ref_dir=os.path.join(base_dir, dataset_dir, '06_normals_pcpnet'), ref_ending='.normals',
dirs=dirs, endings_per_dir=endings_per_dir,
out_dir=os.path.join(base_dir, dataset_dir, '07_pts_normals_pcpnet'), out_ending='.xyz')
print('### poisson reconstruction from pcpnet normals')
apply_meshlab_filter(base_dir=base_dir, dataset_dir=dataset_dir, pts_dir='07_pts_normals_pcpnet',
recon_mesh_dir='06_poisson_rec_pcpnet_normals', num_processes=num_processes,
filter_file='poisson.mlx', meshlabserver_bin=meshlabserver)
end = time.time()
print('SPSR with PCPNet normals took: {}'.format(end - start))
print('### normal estimation and poisson reconstruction pcpnet - hausdorff distance')
new_meshes_dir_abs = os.path.join(base_dir, dataset_dir, '06_poisson_rec_pcpnet_normals')
ref_meshes_dir_abs = os.path.join(base_dir, dataset_dir, '03_meshes')
csv_file = os.path.join(base_dir, dataset_dir, 'comp_poisson_rec_pcpnet_normals.csv')
val_set_file_abs = os.path.join(base_dir, dataset_dir, 'valset.txt')
evaluation.mesh_comparison(new_meshes_dir_abs=new_meshes_dir_abs, ref_meshes_dir_abs=ref_meshes_dir_abs,
num_processes=num_processes, report_name=csv_file,
samples_per_model=10000, dataset_file_abs=val_set_file_abs)
# this works only when GT meshes are available
print('### get ground truth normals for point cloud')
utils.get_pts_normals(base_dir=base_dir, dataset_dir=dataset_dir,
dir_in_pointcloud='04_pts', dir_in_meshes='03_meshes',
dir_out_normals='06_normals', samples_per_model=100000, num_processes=num_processes)
print('### poisson reconstruction from gt normals')
apply_meshlab_filter(base_dir=base_dir, dataset_dir=dataset_dir, pts_dir='06_normals/pts',
recon_mesh_dir='06_poisson_rec_gt_normals', num_processes=num_processes,
filter_file='poisson.mlx', meshlabserver_bin=meshlabserver)
print('### normal estimation and poisson reconstruction gt normals - hausdorff distance')
new_meshes_dir_abs = os.path.join(base_dir, dataset_dir, '06_poisson_rec_gt_normals')
ref_meshes_dir_abs = os.path.join(base_dir, dataset_dir, '03_meshes')
csv_file = os.path.join(base_dir, dataset_dir, 'comp_poisson_rec_gt_normals.csv')
val_set_file_abs = os.path.join(base_dir, dataset_dir, 'valset.txt')
evaluation.mesh_comparison(new_meshes_dir_abs=new_meshes_dir_abs, ref_meshes_dir_abs=ref_meshes_dir_abs,
num_processes=num_processes, report_name=csv_file,
samples_per_model=10000, dataset_file_abs=val_set_file_abs)