EquiFusion / src /utils /store.py
ceccurr's picture
AMASS sequences chosen by the author, previous MoYo sequences, one generated limb on Nymeria, paper colors, notes
c9816d0 verified
Raw History Blame Contribute Delete
2.74 kB
import os
import zarr
import numpy as np
import yaml
class SequenceStorer:
def __init__(self, store_output_path, num_samples, num_sequences, config,if_gt=False):
self.store_output_path = store_output_path
self.num_samples = num_samples
self.config = config
self.if_gt = if_gt
self.pred_length = self.config['pred_length']
if config['if_eval_from60_at50fps']:
self.pred_length = self.pred_length * 5 // 6
os.makedirs(self.store_output_path, exist_ok=True)
self.setup_store_preds()
print(f"Storing output at {store_output_path}: ", num_sequences, 'segments')
@property
def future_shape(self):
return (self.pred_length, self.config['num_joints'], 3)
@property
def past_shape(self):
return (self.config['obs_length'], self.config['num_joints'], 3)
def _setup_store_preds(self):
store_output_path = os.path.join(self.store_output_path, 'output.zarr')
chunk_size = int(10000 // self.num_samples)
self.output_poses = zarr.open(store_output_path, mode='w', shape=(0, self.num_samples, *self.future_shape),
chunks=(chunk_size, self.num_samples, *self.future_shape), dtype=np.float32)
def _setup_store_gt(self):
self.gt_poses = zarr.open(os.path.join(self.store_output_path, 'gt.zarr'), mode='w', shape=(0, *self.future_shape),
chunks=(1000, *self.future_shape), dtype=np.float32)
self.obs_poses = zarr.open(os.path.join(self.store_output_path, 'obs.zarr'), mode='w', shape=(0, *self.past_shape),
chunks=(1000, *self.past_shape), dtype=np.float32)
def setup_store_preds(self):
if self.if_gt:
self._setup_store_gt()
else:
self._setup_store_preds()
self.output_metadata = {'unique_id': [], 'metadata':[]}
def store_batch(self, output, extra, dataset):
if self.if_gt:
gt, obs = output
self.gt_poses.append(dataset.skeleton.if_add_zero_pad_center_hip(gt).cpu().numpy(), axis=0)
self.obs_poses.append(dataset.skeleton.if_add_zero_pad_center_hip(obs).cpu().numpy(), axis=0)
else:
self.output_poses.append(dataset.skeleton.if_add_zero_pad_center_hip(output).cpu().numpy(), axis=0)
self.output_metadata['unique_id'].extend(dataset.unique_sample_string(extra))
self.output_metadata['metadata'].extend(extra['metadata'][dataset.metadata_class_idx])
def finalize_store(self):
with open(os.path.join(self.store_output_path, 'metadata.yaml'), 'w') as outfile:
yaml.dump(self.output_metadata, outfile, default_flow_style=False)