def train(model_dir, sensitivities_file, eval_metric_loss):
# 定义训练和验证时的transforms
train_transforms = transforms.Compose([
transforms.MixupImage(mixup_epoch=250),
transforms.RandomDistort(),
transforms.RandomExpand(),
transforms.RandomCrop(),
transforms.Resize(target_size=608, interp='RANDOM'),
transforms.RandomHorizontalFlip(),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
transforms.Resize(target_size=608, interp='CUBIC'),
transforms.Normalize()
])
# 定义训练和验证所用的数据集
train_dataset = pdx.datasets.VOCDetection(
data_dir='dataset',
file_list='dataset/train_list.txt',
label_list='dataset/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.VOCDetection(data_dir='dataset',
file_list='dataset/val_list.txt',
label_list='dataset/labels.txt',
transforms=eval_transforms)
if model_dir is None:
# 使用imagenet数据集上的预训练权重
pretrain_weights = "IMAGENET"
else:
assert os.path.isdir(model_dir), "Path {} is not a directory".format(
model_dir)
pretrain_weights = model_dir
save_dir = "output/yolov3_mobile"
if sensitivities_file is not None:
if sensitivities_file != 'DEFAULT':
assert os.path.exists(
sensitivities_file), "Path {} not exist".format(
sensitivities_file)
save_dir = "output/yolov3_mobile_prune"
num_classes = len(train_dataset.labels)
model = pdx.det.YOLOv3(num_classes=num_classes)
model.train(num_epochs=400,
train_dataset=train_dataset,
train_batch_size=10,
eval_dataset=eval_dataset,
learning_rate=0.0001,
lr_decay_epochs=[310, 350],
pretrain_weights=pretrain_weights,
save_dir=save_dir,
use_vdl=True,
sensitivities_file=sensitivities_file,
eval_metric_loss=eval_metric_loss)
def __init__(self, model_path, bs=4, thresh=0.9, autoflush=True):
self.model = pdx.load_model(model_path)
self.transform = dT.Compose([dT.Resize(), dT.Normalize()])
self.bs = bs
self.thresh = thresh
self.imgs = []
self.infos = []
self.autoflush = autoflush
with open('MOT20-03.txt', newline='') as csvfile:
reader = csv.reader(csvfile)
for row in reader:
if int(row[0]) not in bboxes:
bboxes[int(row[0])] = []
bboxes[int(row[0])].append(row[1:])
import paddlex as pdx
from paddlex.det import transforms
INTERACTIVE = True
WRITE_CSV = False
USE_YOLO = False
if USE_YOLO:
eval_transforms = transforms.Compose([
transforms.Resize(target_size=512, interp='CUBIC'),
transforms.Normalize(),
])
else:
eval_transforms = transforms.Compose([
transforms.Normalize(),
transforms.ResizeByShort(short_size=800, max_size=1333),
transforms.Padding(coarsest_stride=32),
])
loop_gen = ((pdx.datasets.VOCDetection(
data_dir=base,
file_list=k,
transforms=eval_transforms,
label_list='./data/labels.txt'), v) for k, v in [(
'data/MOT20/train/MOT20-03/manifest.txt', 'MOT20-03.txt'
base = './data/'
import os
import paddlex as pdx
from paddlex.det import transforms
train_transforms = transforms.Compose([
transforms.MixupImage(mixup_epoch=250),
transforms.RandomDistort(),
transforms.RandomExpand(),
transforms.RandomCrop(),
transforms.Resize(target_size=512, interp='RANDOM'),
transforms.RandomHorizontalFlip(),
transforms.Normalize(),
])
eval_transforms = transforms.Compose([
transforms.Resize(target_size=512, interp='CUBIC'),
transforms.Normalize(),
])
train_dataset = pdx.datasets.VOCDetection(data_dir=base,
file_list=os.path.join(
base, 'train.txt'),
label_list='./data/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.VOCDetection(data_dir=base,
file_list=os.path.join(
base, 'valid.txt'),
transforms=eval_transforms,
label_list='./data/labels.txt')
base = './data/'
import os
import paddlex as pdx
from paddlex.det import transforms
train_transforms = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.Normalize(),
transforms.ResizeByShort(short_size=800, max_size=1333),
transforms.Padding(coarsest_stride=32)
])
eval_transforms = transforms.Compose([
transforms.Normalize(),
transforms.ResizeByShort(short_size=800, max_size=1333),
transforms.Padding(coarsest_stride=32), ])
train_dataset = pdx.datasets.VOCDetection(
data_dir=base,
file_list=os.path.join(base, 'train_all_1.txt'),
label_list='./data/labels.txt',
transforms=train_transforms,
shuffle=True)
eval_dataset = pdx.datasets.VOCDetection(
data_dir=base,
file_list=os.path.join(base, 'valid_all_1.txt'),
transforms=eval_transforms,
label_list='./data/labels.txt')
num_classes = len(train_dataset.labels) + 1
print('class num:', num_classes)
parser.add_argument("-o",
"--output",
type=str,
default="/home/aistudio/test/frame",
help="结果帧存放路径")
parser.add_argument("-m",
"--model",
type=str,
default="/home/aistudio/pdx/output/yolov3/best_model",
help="起落架检测模型路径")
args = parser.parse_args()
people_det = hub.Module(name="yolov3_resnet50_vd_coco2017")
flg_det = pdx.load_model(args.model)
transforms = transforms.Compose([transforms.Resize(), transforms.Normalize()])
for vid_name in tqdm(os.listdir(args.input)):
print("processing {}".format(vid_name))
vidcap = cv2.VideoCapture(osp.join(args.input, vid_name))
idx = 0
while True:
vidcap.set(1, idx)
success, image = vidcap.read()
if not success:
break
flg = flg_det.predict(image, transforms=transforms)
print(flg)
if len(flg) == 0:
elif color == "B":
color = (255, 0, 0)
for l in lines:
cv2.line(img, l[0], l[1], color, 2)
cls_model = pdx.load_model('output/gear_clas/')
cls_trans = cls_transforms.Compose([
# cls_transforms.Resize(),
cls_transforms.Normalize()
])
det_model = pdx.load_model('output/gear_det/')
det_trans = det_transforms.Compose([
det_transforms.Resize(),
det_transforms.Normalize()
])
vid_dir = "/home/aistudio/plane/vid-split/train/"
itv = 5
bs = 4
for vid_name in tqdm(os.listdir(vid_dir)):
print(vid_name)
vidcap = cv2.VideoCapture(osp.join(vid_dir, vid_name))
frame_data = []
index = []
frame_idx = 0
success = True
count = 0
# ])
# train_transforms = t.Compose([t.ComposedYOLOv3Transforms("train")])
# eval_transforms = t.Compose([t.ComposedYOLOv3Transforms("eval")])
width = 480
height = 270
mean=[0.485, 0.456, 0.406]
std=[0.229, 0.224, 0.225]
epoch_num = 20
train_transforms = t.Compose([
# t.RandomHorizontalFlip(),
# t.RandomExpand(),
# t.RandomDistort(),
# t.MixupImage(mixup_epoch=int(epoch_num * 0.5)),
# t.Resize(target_size=width, interp='RANDOM'),
t.Normalize(mean=mean, std=std),
])
# 定义训练和验证所用的数据集
# API说明：https://paddlex.readthedocs.io/zh_CN/develop/apis/datasets.html#paddlex-datasets-vocdetection
train_dataset = pdx.datasets.CocoDetection(
data_dir='/home/aistudio/data/data67498/DatasetId_153212_1611125306/Images',
ann_file='/home/aistudio/data/data67498/DatasetId_153212_1611125306/Annotations/coco_info.json',
transforms=train_transforms,
num_workers=8,
buffer_size=256,
parallel_method='process',
shuffle=True)
# eval_dataset = pdx.datasets.CocoDetection(
# data_dir='/home/aistudio/data/data67498/DatasetId_152881_1610856374/Images',
import paddlex as pdx
from paddlex.det import transforms
eval_transforms = transforms.Compose([
transforms.Normalize(),
transforms.ResizeByShort(short_size=800, max_size=1333),
transforms.Padding(coarsest_stride=32), ])
model = pdx.load_model('./FasterRCNN/best_model')
image_name = './test.jpg'
result = model.predict(image_name, eval_transforms)
pdx.det.visualize(image_name, result, threshold=0.1)
ymin, xmin, ymax, xmax = b
lines = [[(xmin, ymin), (xmin, ymax)], [(xmax, ymin), (xmax, ymax)],
[(xmin, ymin), (xmax, ymin)], [(xmin, ymax), (xmax, ymax)]]
if color == "R":
color = (0, 0, 255)
elif color == "G":
color = (0, 255, 0)
elif color == "B":
color = (255, 0, 0)
for l in lines:
cv2.line(img, l[0], l[1], color, 2)
flg_det = pdx.load_model("./model/new_flg/")
flg_trans = transforms.Compose([transforms.Resize(), transforms.Normalize()])
people_det = pdx.load_model("./model/person/50")
people_trans = transforms.Compose(
[transforms.Resize(), transforms.Normalize()])
def sort_key(p):
return (int(p['bbox'][0]), int(p['bbox'][1]))
for vid_name in tqdm(os.listdir(vid_dir)):
print(vid_name)
vidcap = cv2.VideoCapture(osp.join(vid_dir, vid_name))
vid_name = vid_name.split(".")[0]
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import paddlex as pdx
import paddle.fluid as fluid
from paddlex.det import transforms
train_transforms = transforms.Compose([
transforms.MixupImage(alpha=1.5, beta=1.5, mixup_epoch=-1),
transforms.RandomExpand(),
transforms.RandomCrop(),
transforms.Resize(target_size=480),
transforms.RandomHorizontalFlip(prob=0.5),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
#transforms.Resize(target_size=480, interp='RANDOM'),
transforms.Resize(target_size=480),
transforms.Normalize()
])
#读取数据
train_dataset = pdx.datasets.VOCDetection(
data_dir='./dataset',
file_list='./dataset/train_list.txt',
label_list='./dataset/label_list.txt',
transforms=train_transforms)
eval_dataset = pdx.datasets.VOCDetection(data_dir='./dataset',
file_list='./dataset/dev_list.txt',
label_list='./dataset/label_list.txt',
transforms=eval_transforms)
# 3
#6.训练。训练过程中可点击侧边栏的可视化，启动VisualDL，支持可视化训练过程
# 环境变量配置，用于控制是否使用GPU
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
from paddlex.det import transforms
import paddlex as pdx
train_transforms = transforms.Compose([
transforms.RandomDistort(),
transforms.RandomCrop(),
transforms.RandomHorizontalFlip(),
transforms.ResizeByShort(short_size=[800], max_size=1333),
transforms.Normalize(mean=[0.5], std=[0.5]),
transforms.Padding(coarsest_stride=32)
])
eval_transforms = transforms.Compose([
transforms.ResizeByShort(short_size=800, max_size=1333),
transforms.Normalize(),
transforms.Padding(coarsest_stride=32),
])
# 定义训练和验证所用的数据集
train_dataset = pdx.datasets.VOCDetection(
data_dir='D:/code/datasets/cz/tile_round1_train_20201231',
file_list='D:/code/datasets/cz/tile_round1_train_20201231/train.txt',
label_list='D:/code/datasets/cz/tile_round1_train_20201231/labels.txt',
transforms=train_transforms,
# 设置使用0号GPU卡（如无GPU，执行此代码后仍然会使用CPU训练模型）
import matplotlib
matplotlib.use('Agg')
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import paddlex as pdx
from paddlex.det import transforms
train_transforms = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.Padding(coarsest_stride=32),
transforms.RandomDistort(contrast_prob=0.01,
hue_prob=0,
saturation_prob=0.01),
transforms.Normalize()
])
eval_transforms = transforms.Compose([
transforms.RandomDistort(),
transforms.Padding(coarsest_stride=32),
transforms.Normalize()
])
#读取数据
train_dataset = pdx.datasets.VOCDetection(
data_dir='./dataset',
file_list='./dataset/train_list.txt',
label_list='./dataset/label_list.txt',
transforms=train_transforms)
eval_dataset = pdx.datasets.VOCDetection(data_dir='./dataset',
file_list='./dataset/dev_list.txt',
label_list='./dataset/label_list.txt',
from tqdm import tqdm
from deep_sort import DeepSort
from evaluation.CsvEvalWriter import CsvEvalWriter
sort = DeepSort('models/deep_sort/checkpoint_static/net', n_init=2)
base = './data/'
import os
import paddlex as pdx
from paddlex.det import transforms
INTERACTIVE = False
WRITE_CSV = True
eval_transforms = transforms.Compose([
transforms.Resize(target_size=512, interp='CUBIC'),
transforms.Normalize(),
])
eval_dataset = pdx.datasets.VOCDetection(
data_dir=base,
file_list=os.path.join(base, 'valid_all_1.txt'),
label_list='./data/labels.txt')
model = pdx.load_model('./YOLOv3/best_model')
evaluator = CsvEvalWriter()
for i in tqdm(eval_dataset.file_list):
image_name = i[0]
start = time.time()
result = model.predict(image_name)
# print('infer time:{:.6f}s'.format(time.time()-start))
# print('detected num:', len(result))
# paddle.disable_static()
