dataloader __getitem__ supported compressed
这个提交包含在:
8
utils.py
8
utils.py
@ -4,6 +4,7 @@ import os
|
||||
import h5py
|
||||
import pickle
|
||||
import fnmatch
|
||||
import cv2
|
||||
from torch.utils.data import TensorDataset, DataLoader
|
||||
|
||||
import IPython
|
||||
@ -43,6 +44,7 @@ class EpisodicDataset(torch.utils.data.Dataset):
|
||||
is_sim = root.attrs['sim']
|
||||
except:
|
||||
is_sim = False
|
||||
compressed = root.attrs.get('compress', False)
|
||||
if '/base_action' in root:
|
||||
base_action = root['/base_action'][()]
|
||||
base_action = preprocess_base_action(base_action)
|
||||
@ -59,6 +61,12 @@ class EpisodicDataset(torch.utils.data.Dataset):
|
||||
image_dict = dict()
|
||||
for cam_name in self.camera_names:
|
||||
image_dict[cam_name] = root[f'/observations/images/{cam_name}'][start_ts]
|
||||
|
||||
if compressed:
|
||||
for cam_name in image_dict.keys():
|
||||
decompressed_image = cv2.imdecode(image_dict[cam_name], 1)
|
||||
image_dict[cam_name] = np.array(decompressed_image)
|
||||
|
||||
# get all actions after and including start_ts
|
||||
if is_sim:
|
||||
action = action[start_ts:]
|
||||
|
||||
在新工单中引用
屏蔽一个用户