-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdali.py
187 lines (167 loc) · 7.53 KB
/
dali.py
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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
import os
import torch
import types
import collections
import numpy as np
# import cupy as cp
from os.path import join
from random import shuffle
from nvidia.dali.pipeline import Pipeline
from nvidia.dali.plugin.pytorch import DALIGenericIterator
import nvidia.dali.fn as fn
import nvidia.dali.types as types
import nvidia.dali.ops as ops
class D2InputIterator(object):
def __init__(self, ann_root, img_root, target_transform=None, batch_size=8, device_id=0, num_gpus=1, is_val=False):
self.ann_root = ann_root
self.img_root = img_root
self.batch_size = batch_size
self.target_transform = target_transform
subfolders = os.listdir(ann_root)
self.anns = []
for folder in subfolders:
if not is_val and not folder.startswith("0008"):
txt_files = os.listdir(join(ann_root, folder))
self.anns.extend([join(folder, txt) for txt in txt_files])
elif is_val and folder.startswith("0008"):
txt_files = os.listdir(join(ann_root, folder))
self.anns.extend([join(folder, txt) for txt in txt_files])
self.n = len(self.anns)
# 注意: 每个迭代器的读取的 data以及gt 根据使用的gpu 数量而决定，每个gpu读取数据的一部分
self.anns = self.anns[self.n * device_id // num_gpus: self.n * (device_id+1) // num_gpus]
labels = ["__background__",
"car", "van", "bus", "truck", "person", "bicycle", "motorcycle", "open-tricycle",
"closed-tricycle", "forklift", "large-block", "small-block"]
self._classes_names = labels
self.ann_map = {item: i for i, item in enumerate(labels)}
def __iter__(self):
self.i = 0
shuffle(self.anns)
return self
def __len__(self):
return self.n
def __next__(self):
batch = []
labels = []
boxes = []
if self.i >= self.n:
raise StopIteration
for _ in range(self.batch_size):
ann_file = self.anns[self.i]
img_file = ann_file.replace(".xml", ".mp4/img1")
img_file = img_file.replace(".txt", ".jpg")
with open(join(self.ann_root, ann_file), "r") as f:
lines = f.readlines()
img = open(join(self.img_root, img_file), 'rb')
bboxs, ids = [], []
for line in lines:
frame_num, _, label, x, y, w, h = line.rstrip().split(",")
x, y, w, h = list(map(float, [x, y, w, h]))
bboxs.append([x, y, x + w, y + h])
ids.append([self.ann_map[label]])
batch.append(np.frombuffer(img.read(), dtype=np.uint8))
ids = torch.tensor(ids, dtype=torch.uint8)
bboxs = torch.tensor(bboxs, dtype=torch.float32)
bboxs, ids = self.target_transform(bboxs, ids)
labels.append(ids.numpy())
boxes.append(bboxs.numpy())
self.i = (self.i + 1) % self.n
# 注意 batch_size (n,3,h,w), labels_size (n, total_anchors * 1) 1 for class, boxes_size (n, total_anchors, 4) 4 for x,y,w,h, 统一输出格式，不能因为一张图片中检测目标数量的不同，而labels和boxes 的 shope 不同
return (batch, labels, boxes)
next = __next__
class D2Pipeline(Pipeline):
def __init__(self, resize, batch_size, num_threads, device_id, external_data):
super(D2Pipeline, self).__init__(batch_size,
num_threads,
device_id,
seed=12,
exec_async=False,
exec_pipelined=False)
self.input = ops.ExternalSource()
self.input_label = ops.ExternalSource()
self.input_boxes = ops.ExternalSource()
self.decode = ops.ImageDecoder(device="mixed", output_type=types.RGB)
# 自定义的函数只能在cpu上运行
# self.grid = ops.PythonFunction(function=grid2x2, num_outputs=4)
self.resize = ops.Resize(device="gpu",
resize_x=resize,
resize_y=resize,
interp_type=types.INTERP_LINEAR)
self.external_data = external_data
self.iterator = iter(self.external_data)
def define_graph(self):
# 定义graph 包含3个输入，分别叫做 self.jpegs, self.labels 和 self.boxes
# self.images 经过decode 和 resize 输出
# self.labels 和 self.boxes直接输出
# transform 也是在这个里面做的，参考dali 的 transform
self.jpegs = self.input()
self.labels = self.input_label()
self.boxes = self.input_boxes()
images = self.decode(self.jpegs)
images = self.resize(images.gpu())
return (images, self.labels, self.boxes)
def iter_setup(self):
try:
images, labels, boxes = self.iterator.next()
self.feed_input(self.jpegs, images)
self.feed_input(self.labels, labels)
self.feed_input(self.boxes, boxes)
except StopIteration:
self.iterator = iter(self.external_data)
raise StopIteration
def create_dataloder(ann_root,
img_root,
resize,
batch_size,
device_id=0,
num_gpus=1,
num_threads=2,
target_transform=None,
is_val=False):
#1.首先自定义一个迭代器来读取自己的数据
eii = D2InputIterator(ann_root=ann_root,
img_root=img_root,
batch_size=batch_size,
device_id=device_id,
num_gpus=num_gpus,
target_transform=target_transform,
is_val=is_val)
#2.定义一个 管道 继承于dali 的 pipeline 类
pipe = D2Pipeline(resize=resize,
batch_size=batch_size,
num_threads=num_threads,
device_id=device_id,
external_data=eii)
pii = DALIGenericIterator(pipe,
output_map=["img","labels","boxes"],
size=len(eii),
dynamic_shape=True,
last_batch_padded=True,
fill_last_batch=False)
return pii
if __name__ == '__main__':
batch_size = 32
num_gpus = 1
num_threads = 8
epochs = 1
img_root = "/VisualGroup/share/data/didi/mot_format_didi_data/"
ann_root = "/root/D2DetectAnn/"
# create_dataloader 返回 DALIGenericIterator 迭代器，用法和 pytorch 的dataloader一样，见 174 行
pii = create_dataloder(ann_root,
img_root,
resize=360,
batch_size=4,
device_id=0,
num_gpus=1,
num_threads=1,
is_val=True)
for e in range(epochs):
for i, data in enumerate(pii):
imgs = data[0]["img"]
labels = data[0]["labels"]
boxes = data[0]["boxes"]
print("epoch: {}, iter {}".format(e, i), imgs.shape, labels.shape)
pii.reset()
#######
#主要参考：https://blog.csdn.net/weixin_42028608/article/details/105564060
#######