EquiFusion / src /data /loaders /base /create_dataset_utils.py
ceccurr's picture
EquiFusion demo
e6d357a verified
Raw History Blame Contribute Delete
4.99 kB
import os
import numpy as np
import json
import torch
from torch.utils.data import DataLoader
from functools import partial
from .math_utils import compute_mean_motions, get_multimodal_gt
from .motion_dataset import MotionDataset
def compute_stats(precomputed_folder, Skeleton, Dataset: MotionDataset, if_consider_hip:bool, batch_size = 128, overwrite=False):
np.random.seed(0)
torch.manual_seed(0)
torch.cuda.manual_seed(0)
skeleton = Skeleton[0](if_consider_hip=if_consider_hip)
dataset = Dataset( split='test', skeleton=skeleton, precomputed_folder=precomputed_folder,
if_consider_hip=if_consider_hip, augmentation=0, da_mirroring=0.0, da_rotations=0.0,
dtype="float32")
dest_file = os.path.join(precomputed_folder, "mean_motion_test.txt")
if not os.path.exists(dest_file) or overwrite:
class_average_3d, list_of_motions_3d, frequencies = compute_mean_motions(dataset, batch_size=batch_size)
with open(dest_file, 'w') as filehandle:
filehandle.write('\n'.join([f"{c},{meanmot},{freq}" for c, meanmot, freq in zip(list(class_average_3d.keys()), list_of_motions_3d, frequencies)]))
else:
print("Files for Mean motion values already exist. Skipping.")
if if_consider_hip == True: # dataset must deliver items with hip
dest_file = os.path.join(precomputed_folder, "mean_motion_traj_test.txt")
if not os.path.exists(dest_file) or overwrite:
class_average_3d, list_of_motions_3d, frequencies = compute_mean_motions(dataset, batch_size=batch_size, if_traj_only=True)
with open(dest_file, 'w') as filehandle:
filehandle.write('\n'.join([f"{c},{meanmot},{freq}" for c, meanmot, freq in zip(list(class_average_3d.keys()), list_of_motions_3d, frequencies)]))
else:
print("Files for Mean motion values for MOTPRED task - hip - already exist. Skipping.")
# compute stats for centered pose
if_consider_hip = False
dest_file = os.path.join(precomputed_folder, "mean_motion_pose_test.txt")
if not os.path.exists(dest_file) or overwrite:
skeleton = Skeleton[1](if_consider_hip=if_consider_hip)
dataset = Dataset(split='test', skeleton=skeleton, precomputed_folder=precomputed_folder, if_consider_hip=if_consider_hip,
augmentation=0, da_mirroring=0.0, da_rotations=0.0,
dtype="float32")
class_average_3d, list_of_motions_3d, frequencies = compute_mean_motions(dataset, batch_size=batch_size)
with open(dest_file, 'w') as filehandle:
filehandle.write('\n'.join([f"{c},{meanmot},{freq}" for c, meanmot, freq in zip(list(class_average_3d.keys()), list_of_motions_3d, frequencies)]))
else:
print("Files for Mean motion values for MOTPRED task - pose - already exist. Skipping.")
def compute_multimodal_gt(Dataset: MotionDataset,
annotations_folder: str, precomputed_folder: str, split='test',
multimodal_threshold=0.5, segment_path="/hmp/segments_0.5s-2s", batch_size=2048):
mmgt_path = os.path.join(precomputed_folder, f"mmgt_{split}.txt")
if os.path.exists(mmgt_path):
print("Multimodal GT already exists. Skipping.")
return
mmgt = compute_multimodal_gt_onsplit(Dataset=partial(Dataset, split=split, segments_path = segment_path,
annotations_folder=annotations_folder, precomputed_folder=precomputed_folder),
dest_path=mmgt_path,
multimodal_threshold=multimodal_threshold, batch_size=batch_size)
import ast
with open(os.path.join(precomputed_folder, f"mmgt_{split}.txt"), 'r') as filehandle:
mmgt_loaded = ast.literal_eval(json.load(filehandle))
assert mmgt_loaded==mmgt, "Loaded multimodal GT is different from the original one."
def compute_multimodal_gt_onsplit(Dataset: MotionDataset,
dest_path: str,
multimodal_threshold=0.5, batch_size=2048):
np.random.seed(0)
torch.manual_seed(0)
torch.cuda.manual_seed(0)
dataset = Dataset(#annotations_folder=annotations_folder, precomputed_folder=precomputed_folder,
augmentation=0, da_mirroring=0.0, da_rotations=0.0, dtype="float32")
data_loader = DataLoader(dataset, pin_memory= True, shuffle=False, batch_size=batch_size,
num_workers=0, drop_last=False)
mmgt = get_multimodal_gt(data_loader, multimodal_threshold=multimodal_threshold, recover_landmarks=True)#, store_tempfile=dest_path if 'train' in dest_path else None) # recover_landmarks makes no difference for CenterPoseSkeleton
# analyze and store
with open(dest_path, 'w') as filehandle:
json.dump(str(mmgt), filehandle)
return mmgt