def evaluate(experiment_directory, checkpoint, data_dir, split_filename):
with open(split_filename, "r") as f:
split = json.load(f)
chamfer_results = []
p = Pool(8)
ds = []
cn = []
inn = []
exd = []
ckp = []
dtd = []
print('data_preparing')
for dataset in split:
for class_name in split[dataset]:
for iii, instance_name in enumerate(split[dataset][class_name]):
if iii >= 300:
break
ds.append(dataset)
cn.append(class_name)
inn.append(instance_name)
exd.append(experiment_directory)
ckp.append(checkpoint)
dtd.append(data_dir)
print('multi thread start')
chamfer_results = p.map(evaluate_one_instance, ds, cn, inn, exd, ckp, dtd)
print(np.mean([q[1] for q in chamfer_results]), np.median([q[1] for q in chamfer_results]))
print(np.mean([q[2] for q in chamfer_results]), np.median([q[2] for q in chamfer_results]))
with open(
os.path.join(
ws.get_evaluation_dir(experiment_directory, checkpoint, True), "chamfer_and_emd.csv"
),
"w",
) as f:
f.write("shape, chamfer_dist\n")
for result in chamfer_results:
f.write("{}, {}, {}\n".format(result[0], result[1], result[2]))
def evaluate(experiment_directory, checkpoint, data_dir, split_filename):
with open(split_filename, "r") as f:
split = json.load(f)
chamfer_results = []
for dataset in split:
for class_name in split[dataset]:
for instance_name in split[dataset][class_name]:
logging.debug("evaluating " +
os.path.join(dataset, class_name, instance_name))
reconstructed_mesh_filename = ws.get_reconstructed_mesh_filename(
experiment_directory, checkpoint, dataset, class_name,
instance_name)
logging.debug('reconstructed mesh is "' +
reconstructed_mesh_filename + '"')
ground_truth_samples_filename = os.path.join(
data_dir,
"SurfaceSamples",
dataset,
class_name,
instance_name + ".npy",
)
logging.debug("ground truth samples are " +
ground_truth_samples_filename)
# normalization_params_filename = os.path.join(
# data_dir,
# "NormalizationParameters",
# dataset,
# class_name,
# instance_name + ".npz",
# )
# logging.debug(
# "normalization params are " + ground_truth_samples_filename
# )
ground_truth_points = np.load(ground_truth_samples_filename)
ground_truth_points = ground_truth_points.reshape(-1, 2)
reconstruction = np.load(reconstructed_mesh_filename)
reconstruction = reconstruction.reshape(-1, 2)
# normalization_params = np.load(normalization_params_filename)
chamfer_dist = deep_sdf.metrics.chamfer.compute_trimesh_chamfer(
ground_truth_points, reconstruction)
logging.debug("chamfer distance: " + str(chamfer_dist))
chamfer_results.append(
(os.path.join(dataset, class_name,
instance_name), chamfer_dist))
with open(
os.path.join(
ws.get_evaluation_dir(experiment_directory, checkpoint, True),
"chamfer.csv"),
"w",
) as f:
f.write("shape, chamfer_dist\n")
for result in chamfer_results:
f.write("{}, {}\n".format(result[0], result[1]))