Spaces:
Running
Running
File size: 3,315 Bytes
29f689c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 |
import numbers
from collections import defaultdict
import numpy as np
import torch
class DictCollator(object):
"""data batch."""
def __call__(self, batch):
data_dict = defaultdict(list)
to_tensor_keys = []
for sample in batch:
for k, v in sample.items():
if isinstance(v, (np.ndarray, torch.Tensor, numbers.Number)):
if k not in to_tensor_keys:
to_tensor_keys.append(k)
data_dict[k].append(v)
for k in to_tensor_keys:
data_dict[k] = torch.from_numpy(data_dict[k])
return data_dict
class ListCollator(object):
"""data batch."""
def __call__(self, batch):
data_dict = defaultdict(list)
to_tensor_idxs = []
for sample in batch:
for idx, v in enumerate(sample):
if isinstance(v, (np.ndarray, torch.Tensor, numbers.Number)):
if idx not in to_tensor_idxs:
to_tensor_idxs.append(idx)
data_dict[idx].append(v)
for idx in to_tensor_idxs:
data_dict[idx] = torch.from_numpy(data_dict[idx])
return list(data_dict.values())
class SSLRotateCollate(object):
"""
bach: [
[(4*3xH*W), (4,)]
[(4*3xH*W), (4,)]
...
]
"""
def __call__(self, batch):
output = [np.concatenate(d, axis=0) for d in zip(*batch)]
return output
class DyMaskCollator(object):
"""
batch: [
image [batch_size, channel, maxHinbatch, maxWinbatch]
image_mask [batch_size, channel, maxHinbatch, maxWinbatch]
label [batch_size, maxLabelLen]
label_mask [batch_size, maxLabelLen]
...
]
"""
def __call__(self, batch):
max_width, max_height, max_length = 0, 0, 0
bs, channel = len(batch), batch[0][0].shape[0]
proper_items = []
for item in batch:
if item[0].shape[1] * max_width > 1600 * 320 or item[0].shape[
2] * max_height > 1600 * 320:
continue
max_height = item[0].shape[
1] if item[0].shape[1] > max_height else max_height
max_width = item[0].shape[
2] if item[0].shape[2] > max_width else max_width
max_length = len(
item[1]) if len(item[1]) > max_length else max_length
proper_items.append(item)
images, image_masks = np.zeros(
(len(proper_items), channel, max_height, max_width),
dtype='float32'), np.zeros(
(len(proper_items), 1, max_height, max_width), dtype='float32')
labels, label_masks = np.zeros((len(proper_items), max_length),
dtype='int64'), np.zeros(
(len(proper_items), max_length),
dtype='int64')
for i in range(len(proper_items)):
_, h, w = proper_items[i][0].shape
images[i][:, :h, :w] = proper_items[i][0]
image_masks[i][:, :h, :w] = 1
l = len(proper_items[i][1])
labels[i][:l] = proper_items[i][1]
label_masks[i][:l] = 1
return images, image_masks, labels, label_masks
|