-
Notifications
You must be signed in to change notification settings - Fork 85
/
train.py
195 lines (175 loc) · 9.04 KB
/
train.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
188
189
190
191
192
193
194
195
import sys
sys.path.append('../yolov5_in_tf2_keras')
import os
import tqdm
import numpy as np
import random
import tensorflow as tf
from data.visual_ops import draw_bounding_box
from data.generate_coco_data import CoCoDataGenrator
from yolo import Yolo
from loss import ComputeLoss
from val import val
from layers import nms
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
def main():
epochs = 300
log_dir = './logs'
# 可以选择 ['5l', '5s', '5m', '5x']
yolov5_type = "5s"
image_shape = (320, 320, 3)
# num_class = 91
num_class = 2
batch_size = 20
# -1表示全部数据参与训练
train_img_nums = -1
train_coco_json = './data/cat_dog_face_data/train_annotations.json'
val_coco_json = './data/cat_dog_face_data/val_annotations.json'
# 类别名, 也可以自己提供一个数组, 不通过coco
# classes = ['none', 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
# 'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'none', 'stop sign',
# 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant',
# 'bear', 'zebra', 'giraffe', 'none', 'backpack', 'umbrella', 'none', 'none', 'handbag',
# 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat',
# 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle', 'none', 'wine glass',
# 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange', 'broccoli',
# 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed', 'none',
# 'dining table', 'none', 'none', 'toilet', 'none', 'tv', 'laptop', 'mouse', 'remote', 'keyboard',
# 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'none', 'book', 'clock',
# 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush']
classes = ['cat', 'dog']
# 这里anchor归一化到[0,1]区间
anchors = np.array([[10, 13], [16, 30], [33, 23],
[30, 61], [62, 45], [59, 119],
[116, 90], [156, 198], [373, 326]]) / 640.
anchors = np.array(anchors, dtype=np.float32)
# 分别对应1/8, 1/16, 1/32预测输出层
anchor_masks = np.array([[0, 1, 2], [3, 4, 5], [6, 7, 8]], dtype=np.int8)
# tensorboard日志
summary_writer = tf.summary.create_file_writer(log_dir)
# data generator
coco_data = CoCoDataGenrator(
coco_annotation_file= train_coco_json,
train_img_nums=train_img_nums,
img_shape=image_shape,
batch_size=batch_size,
include_mask=False,
include_crowd=False,
include_keypoint=False,
need_down_image=False,
using_argument=True
)
# 验证集
val_coco_data = CoCoDataGenrator(
coco_annotation_file=val_coco_json,
train_img_nums=-1,
img_shape=image_shape,
batch_size=batch_size,
include_mask=False,
include_crowd=False,
include_keypoint=False,
need_down_image=False,
using_argument=False
)
yolo = Yolo(
image_shape=image_shape,
batch_size=batch_size,
num_class=num_class,
is_training=True,
anchors=anchors,
anchor_masks=anchor_masks,
net_type=yolov5_type
)
yolo.yolov5.summary(line_length=200)
optimizer = tf.keras.optimizers.Adam(learning_rate=0.0001)
loss_fn = ComputeLoss(
image_shape=image_shape,
anchors=anchors,
anchor_masks=anchor_masks,
num_class=num_class,
anchor_ratio_thres=4,
only_best_anchor=False,
balanced_rate=15,
iou_ignore_thres=0.5
)
pre_mAP = 0.
# data = coco_data.next_batch()
for epoch in range(epochs):
train_progress_bar = tqdm.tqdm(range(coco_data.total_batch_size), desc="train epoch {}/{}".format(epoch, epochs-1), ncols=100)
for batch in train_progress_bar:
with tf.GradientTape() as tape:
data = coco_data.next_batch()
valid_nums = data['valid_nums']
gt_imgs = np.array(data['imgs'] / 255., dtype=np.float32)
gt_boxes = np.array(data['bboxes'] / image_shape[0], dtype=np.float32)
gt_classes = data['labels']
# print("-------epoch {}, step {}, total step {}--------".format(epoch, batch,
# epoch * coco_data.total_batch_size + batch))
# print("current data index: ",
# coco_data.img_ids[(coco_data.current_batch_index - 1) * coco_data.batch_size:
# coco_data.current_batch_index * coco_data.batch_size])
# for i, nums in enumerate(valid_nums):
# print("gt boxes: ", gt_boxes[i, :nums, :] * image_shape[0])
# print("gt classes: ", gt_classes[i, :nums])
yolo_preds = yolo.yolov5(gt_imgs, training=True)
loss_xy, loss_wh, loss_box, loss_obj, loss_cls = loss_fn(yolo_preds, gt_boxes, gt_classes)
total_loss = loss_box + loss_obj + loss_cls
train_progress_bar.set_postfix(ordered_dict={"loss":'{:.5f}'.format(total_loss)})
grad = tape.gradient(total_loss, yolo.yolov5.trainable_variables)
optimizer.apply_gradients(zip(grad, yolo.yolov5.trainable_variables))
# Scalar
with summary_writer.as_default():
tf.summary.scalar('loss/box_loss', loss_box,
step=epoch * coco_data.total_batch_size + batch)
tf.summary.scalar('loss/object_loss', loss_obj,
step=epoch * coco_data.total_batch_size + batch)
tf.summary.scalar('loss/class_loss', loss_cls,
step=epoch * coco_data.total_batch_size + batch)
tf.summary.scalar('loss/total_loss', total_loss,
step=epoch * coco_data.total_batch_size + batch)
# image, 只拿每个batch的其中一张
random_one = random.choice(range(batch_size))
# gt
gt_img = gt_imgs[random_one].copy() * 255
gt_box = gt_boxes[random_one] * image_shape[0]
gt_class = gt_classes[random_one]
non_zero_ids = np.where(np.sum(gt_box, axis=-1))[0]
for i in non_zero_ids:
cls = gt_class[i]
class_name = coco_data.coco.cats[cls]['name']
xmin, ymin, xmax, ymax = gt_box[i]
# print(xmin, ymin, xmax, ymax)
gt_img = draw_bounding_box(gt_img, class_name, cls, int(xmin), int(ymin), int(xmax), int(ymax))
# pred, 同样只拿第一个batch的pred
pred_img = gt_imgs[random_one].copy() * 255
yolo_head_output = yolo.yolo_head(yolo_preds, is_training=False)
nms_output = nms(image_shape, yolo_head_output.numpy(), iou_thres=0.3)
if len(nms_output) == batch_size:
nms_output = nms_output[random_one]
for box_obj_cls in nms_output:
if box_obj_cls[4] > 0.5:
label = int(box_obj_cls[5])
if coco_data.coco.cats.get(label):
class_name = coco_data.coco.cats[label]['name']
# class_name = classes[label]
xmin, ymin, xmax, ymax = box_obj_cls[:4]
pred_img = draw_bounding_box(pred_img, class_name, box_obj_cls[4], int(xmin), int(ymin),
int(xmax), int(ymax))
concat_imgs = tf.concat([gt_img[:, :, ::-1], pred_img[:, :, ::-1]], axis=1)
summ_imgs = tf.expand_dims(concat_imgs, 0)
summ_imgs = tf.cast(summ_imgs, dtype=tf.uint8)
with summary_writer.as_default():
tf.summary.image("imgs/gt,pred,epoch{}".format(epoch), summ_imgs,
step=epoch * coco_data.total_batch_size + batch)
# 这里计算一下训练集的mAP
val(model=yolo, val_data_generator=coco_data, classes=classes, desc='training dataset val')
# 这里计算验证集的mAP
mAP50, mAP, final_df = val(model=yolo, val_data_generator=val_coco_data, classes=classes, desc='val dataset val')
if mAP > pre_mAP:
pre_mAP = mAP
yolo.yolov5.save_weights(log_dir+"/yolov{}-best.h5".format(yolov5_type))
print("save {}/yolov{}-best.h5 best weight with {} mAP.".format(log_dir, yolov5_type, mAP))
yolo.yolov5.save_weights(log_dir+"/yolov{}-last.h5".format(yolov5_type))
print("save {}/yolov{}-last.h5 last weights at epoch {}.".format(log_dir, yolov5_type, epoch))
if __name__ == "__main__":
main()