def get_ds(cat_desc, regime='e'):
view_index = 5
edge_length_threshold = 0.02
cat_id = cat_desc_to_id(cat_desc)
model_id = '%s_%s' % (regime, cat_desc)
image_ds = RenderConfig().get_dataset(cat_id, view_index)
cloud_ds = get_cloud_manager(
model_id, pre_sampled=True, n_samples=n_samples).get_lazy_dataset()
mesh_ds = get_inferred_mesh_dataset(
model_id, edge_length_threshold=edge_length_threshold)
gt_mesh_ds = get_mesh_dataset(cat_id)
voxel_ds = get_voxel_dataset(
model_id, edge_length_threshold=edge_length_threshold, filled=False)
selected_template_ds = get_selected_template_idx_dataset(model_id)
template_meshes = []
with gt_mesh_ds:
for template_id in get_template_ids(cat_id):
mesh = gt_mesh_ds[template_id]
template_meshes.append(
{k: np.array(mesh[k]) for k in ('vertices', 'faces')})
template_mesh_ds = selected_template_ds.map(lambda i: template_meshes[i])
return Dataset.zip(
image_ds, gt_mesh_ds, cloud_ds, mesh_ds, voxel_ds, template_mesh_ds)
def get_ds(cat_desc, regime='e'):
view_index = 5
edge_length_threshold = 0.02
cat_id = cat_desc_to_id(cat_desc)
model_id = '%s_%s' % (regime, cat_desc)
image_ds = RenderConfig().get_dataset(cat_id, view_index)
cloud_ds = get_cloud_manager(
model_id, pre_sampled=True, n_samples=n_samples).get_lazy_dataset()
mesh_ds = get_inferred_mesh_dataset(
model_id, edge_length_threshold=edge_length_threshold)
gt_mesh_ds = get_mesh_dataset(cat_id)
voxel_ds = get_voxel_dataset(
model_id, edge_length_threshold=edge_length_threshold, filled=False)
selected_template_ds = get_selected_template_idx_dataset(model_id)
template_meshes = []
with gt_mesh_ds:
for template_id in get_template_ids(cat_id):
mesh = gt_mesh_ds[template_id]
template_meshes.append(
{k: np.array(mesh[k]) for k in ('vertices', 'faces')})
template_mesh_ds = selected_template_ds.map(lambda i: template_meshes[i])
return Dataset.zip(
image_ds, gt_mesh_ds, cloud_ds, mesh_ds, voxel_ds, template_mesh_ds)
def create_split_mesh(
cat_desc, edge_length_threshold, overwrite=False,
start_threshold=None):
"""Create split mesh data for templates."""
from shapenet.core import cat_desc_to_id
from shapenet.core.meshes.config import get_mesh_config
from template_ffd.templates.ids import get_template_ids
cat_id = cat_desc_to_id(cat_desc)
example_ids = get_template_ids(cat_id)
config = get_mesh_config(edge_length_threshold)
init = None if start_threshold is None else get_mesh_config(
start_threshold)
config.create_cat_data(cat_id, example_ids, overwrite, init)
def create_split_mesh(cat_desc,
edge_length_threshold,
overwrite=False,
start_threshold=None):
"""Create split mesh data for templates."""
from shapenet.core import cat_desc_to_id
from shapenet.core.meshes.config import get_mesh_config
from template_ffd.templates.ids import get_template_ids
cat_id = cat_desc_to_id(cat_desc)
example_ids = get_template_ids(cat_id)
config = get_mesh_config(edge_length_threshold)
init = None if start_threshold is None else get_mesh_config(
start_threshold)
config.create_cat_data(cat_id, example_ids, overwrite, init)
def create_split(self, cat_id, overwrite=False):
import random
from shapenet.core import get_example_ids
from template_ffd.templates.ids import get_template_ids
if not overwrite and self.has_split(cat_id):
return
template_ids = set(get_template_ids(cat_id))
example_ids = get_example_ids(cat_id)
example_ids = [i for i in example_ids if i not in template_ids]
example_ids.sort()
random.seed(self._seed)
random.shuffle(example_ids)
train_ids, eval_ids = _train_eval_partition(example_ids,
self._train_prop)
train_ids.sort()
eval_ids.sort()
for mode, ids in (('train', train_ids), ('eval', eval_ids)):
with open(self.get_txt_path(cat_id, mode), 'w') as fp:
fp.writelines(('%s\n' % i for i in ids))
def create_split(self, cat_id, overwrite=False):
import random
from shapenet.core import get_example_ids
from template_ffd.templates.ids import get_template_ids
if not overwrite and self.has_split(cat_id):
return
template_ids = set(get_template_ids(cat_id))
example_ids = get_example_ids(cat_id)
example_ids = [i for i in example_ids if i not in template_ids]
example_ids.sort()
random.seed(self._seed)
random.shuffle(example_ids)
train_ids, eval_ids = _train_eval_partition(
example_ids, self._train_prop)
train_ids.sort()
eval_ids.sort()
for mode, ids in (('train', train_ids), ('eval', eval_ids)):
with open(self.get_txt_path(cat_id, mode), 'w') as fp:
fp.writelines(('%s\n' % i for i in ids))
def get_hist_data(model_id, n_bins, mode):
from shapenet.core import cat_desc_to_id
from template_ffd.templates.ids import get_template_ids
from template_ffd.model import load_params
from template_ffd.inference.predictions import get_predictions_dataset
cat_id = cat_desc_to_id(load_params(model_id)['cat_desc'])
n_templates = len(get_template_ids(cat_id))
counts = np.zeros((n_bins, ), dtype=np.int32)
argmax_counts = np.zeros((n_templates, ), dtype=np.int32)
with get_predictions_dataset(model_id) as dataset:
for example_id in dataset:
probs = np.array(dataset[example_id]['probs'])
counts[int(np.max(probs) * n_bins)] += 1
# prob_indices = np.array(0.999*probs * n_bins, dtype=np.int32)
# for pi in prob_indices:
# counts[pi] += 1
argmax_counts[np.argmax(probs)] += 1
counts = counts / np.sum(counts)
argmax_counts = argmax_counts / np.sum(argmax_counts)
return counts, argmax_counts
def template_ids(self):
template_ids = get_template_ids(self.cat_id)
idxs = self.params.get('template_idxs')
if idxs:
template_ids = tuple(template_ids[i] for i in idxs)
return template_ids
def _get_cat_template_ids(cat_id, template_idxs):
from template_ffd.templates.ids import get_template_ids
template_ids = get_template_ids(cat_id)
if template_idxs is not None:
template_ids = tuple(template_ids[i] for i in template_idxs)
return tuple((cat_id, t) for t in template_ids)