def __init__(self, name):
super(Input_Dataset, self).__init__()
# size match
desired_size = (cfg.TRAIN.SEARCH_SIZE - cfg.TRAIN.EXEMPLAR_SIZE) / \
cfg.ANCHOR.STRIDE + 1 + cfg.TRAIN.BASE_SIZE
if desired_size != cfg.TRAIN.OUTPUT_SIZE:
raise Exception('size not match!')
# create anchor target
self.anchor_target = AnchorTarget()
# create datasets
self.dataset = []
start_video = 0
self.num = 0 # 图片序列数目
# create train datasets
if name == 'train':
train_dataset = VID_Dataset(
name,
cfg.DATASET.VID.ROOT,
cfg.DATASET.VID.ANNO,
start_video
)
self.dataset = train_dataset
self.num += train_dataset.num
train_dataset.log()
# create val datasets
if name == 'val':
val_dataset = VID_Dataset(
name,
cfg.DATASET.VID.ROOT,
cfg.DATASET.VID.VALANNO,
start_video
)
self.dataset = val_dataset
self.num += val_dataset.num
val_dataset.log()
# data augmentation
self.template_aug = Augmentation(
cfg.DATASET.TEMPLATE.SHIFT,
cfg.DATASET.TEMPLATE.SCALE,
cfg.DATASET.TEMPLATE.BLUR,
cfg.DATASET.TEMPLATE.FLIP,
cfg.DATASET.TEMPLATE.COLOR
)
self.search_aug = Augmentation(
cfg.DATASET.SEARCH.SHIFT,
cfg.DATASET.SEARCH.SCALE,
cfg.DATASET.SEARCH.BLUR,
cfg.DATASET.SEARCH.FLIP,
cfg.DATASET.SEARCH.COLOR
)
def __init__(self,):
super(TrkDataset, self).__init__()
desired_size = (cfg.TRAIN.SEARCH_SIZE - cfg.TRAIN.EXEMPLAR_SIZE) / \
cfg.ANCHOR.STRIDE + 1 + cfg.TRAIN.BASE_SIZE
if desired_size != cfg.TRAIN.OUTPUT_SIZE:
raise Exception('size not match!')
# create anchor target
self.anchor_target = AnchorTarget()
# create sub dataset
self.all_dataset = []
start = 0
self.num = 0
print()
for name in cfg.DATASET.NAMES:
# print(name)
subdata_cfg = getattr(cfg.DATASET, name)
sub_dataset = SubDataset(
name,
subdata_cfg.ROOT,
subdata_cfg.ANNO,
subdata_cfg.FRAME_RANGE,
subdata_cfg.NUM_USE,
start
)
start += sub_dataset.num
self.num += sub_dataset.num_use
sub_dataset.log()
self.all_dataset.append(sub_dataset)
# data augmentation
self.template_aug = Augmentation(
cfg.DATASET.TEMPLATE.SHIFT,
cfg.DATASET.TEMPLATE.SCALE,
cfg.DATASET.TEMPLATE.BLUR,
cfg.DATASET.TEMPLATE.FLIP,
cfg.DATASET.TEMPLATE.COLOR
)
self.search_aug = Augmentation(
cfg.DATASET.SEARCH.SHIFT,
cfg.DATASET.SEARCH.SCALE,
cfg.DATASET.SEARCH.BLUR,
cfg.DATASET.SEARCH.FLIP,
cfg.DATASET.SEARCH.COLOR
)
videos_per_epoch = cfg.DATASET.VIDEOS_PER_EPOCH
self.num = videos_per_epoch if videos_per_epoch > 0 else self.num
self.num *= cfg.TRAIN.EPOCH
self.pick = self.shuffle()
def __init__(self, model):
super(SiamRPNLatentTracker, self).__init__()
self.score_size = (cfg.TRACK.INSTANCE_SIZE - cfg.TRACK.EXEMPLAR_SIZE) // \
cfg.ANCHOR.STRIDE + 1 + cfg.TRACK.BASE_SIZE
self.anchor_num = len(cfg.ANCHOR.RATIOS) * len(cfg.ANCHOR.SCALES)
hanning = np.hanning(self.score_size)
window = np.outer(hanning, hanning)
self.window = np.tile(window.flatten(), self.anchor_num)
self.anchors = self.generate_anchor(self.score_size)
self.model = model
self.model.eval()
# create anchor target
self.anchor_target = AnchorTarget()
def __init__(self, seq_input_len=3, seq_output_len=1):
super(SeqTrkDataset, self).__init__()
#特征图输出size为：（搜索区域尺寸 -- 模板尺寸)/stride +1 ,没有padding
desired_size = (cfg.TRAIN.SEARCH_SIZE - cfg.TRAIN.EXEMPLAR_SIZE) / \
cfg.ANCHOR.STRIDE + 1 + cfg.TRAIN.BASE_SIZE
if desired_size != cfg.TRAIN.OUTPUT_SIZE:
raise Exception('size not match!')
# create anchor target（初始化事为每个点设置anchor的形状，调用的时候确定哪些anchor是正样本，哪些anchor是负样本）
self.anchor_target = AnchorTarget()
#lstm,gru中连续seq_input_len个tensor输入作为历史信息的学习，后面再输入一个tensor，输出一个状态，后面输入的
#tensor的总个数为个
self.seq_input_len = seq_input_len #lstm ,gru输入序列的长度
self.seq_output_len = seq_output_len #lstm,gru 输出序列的长度
# create sub dataset
self.all_dataset = []
start = 0
self.num = 0 #视频片段的总个数
for name in cfg.DATASET.NAMES:
subdata_cfg = getattr(cfg.DATASET, name)
sub_dataset = SubDataset(name, subdata_cfg.ROOT, subdata_cfg.ANNO,
subdata_cfg.FRAME_RANGE,
subdata_cfg.NUM_USE, start)
start += sub_dataset.num
self.num += sub_dataset.num_use
sub_dataset.log()
self.all_dataset.append(sub_dataset)
# data augmentation（数据增强初始化参数，对于模板的数据增强参数，和搜索区域的数据增强参数不一样，例如模板平移范围为4个像素，而搜索区域平移范围为64个像素）
self.template_aug = Augmentation(cfg.DATASET.TEMPLATE.SHIFT,
cfg.DATASET.TEMPLATE.SCALE,
cfg.DATASET.TEMPLATE.BLUR,
cfg.DATASET.TEMPLATE.FLIP,
cfg.DATASET.TEMPLATE.COLOR)
self.search_aug = Augmentation(cfg.DATASET.SEARCH.SHIFT,
cfg.DATASET.SEARCH.SCALE,
cfg.DATASET.SEARCH.BLUR,
cfg.DATASET.SEARCH.FLIP,
cfg.DATASET.SEARCH.COLOR)
videos_per_epoch = cfg.DATASET.VIDEOS_PER_EPOCH
self.num = videos_per_epoch if videos_per_epoch > 0 else self.num
self.num *= cfg.TRAIN.EPOCH
self.pick = self.shuffle()
self.img_cnt = 0 # 调试用
def __init__(self, cfg):
self.cfg = cfg
# augmentations:
# shift, scale, blur, flip, grayscale
template_augmentors = [
ColorJitterAugmentor(),
ShiftScaleAugmentor(self.cfg.DATASET.TEMPLATE.PAD_RATIO,
self.cfg.DATASET.TEMPLATE.SHIFT,
self.cfg.DATASET.TEMPLATE.SCALE,
self.cfg.DATASET.TEMPLATE.ASPECT,
self.cfg.TRAIN.EXEMPLAR_SIZE,
self.cfg.PREPROC.PIXEL_MEAN[::-1]),
# ResizeAugmentor(self.cfg.TRAIN.EXEMPLAR_SIZE),
]
if self.cfg.DATASET.TEMPLATE.BLUR:
template_augmentors.append(
MotionBlurAugmentor(self.cfg.DATASET.TEMPLATE.BLUR))
# if cfg.DATASET.GRAY:
# template_augmentors.append(GrayscaleAugmentor(cfg.DATASET.GRAY))
self.template_aug = imgaug.AugmentorList(template_augmentors)
search_augmentors = [
ColorJitterAugmentor(),
ShiftScaleAugmentor(self.cfg.DATASET.SEARCH.PAD_RATIO,
self.cfg.DATASET.SEARCH.SHIFT,
self.cfg.DATASET.SEARCH.SCALE,
self.cfg.DATASET.SEARCH.ASPECT,
self.cfg.TRAIN.SEARCH_SIZE,
self.cfg.PREPROC.PIXEL_MEAN[::-1]),
# ResizeAugmentor(self.cfg.TRAIN.SEARCH_SIZE),
]
if self.cfg.DATASET.SEARCH.BLUR:
search_augmentors.append(
MotionBlurAugmentor(self.cfg.DATASET.SEARCH.BLUR))
# if cfg.DATASET.GRAY:
# template_augmentors.append(GrayscaleAugmentor(cfg.DATASET.GRAY))
self.search_aug = imgaug.AugmentorList(search_augmentors)
self.anchor_target = AnchorTarget()
def __init__(self, name, loader, root, frame_root):
super(TrackingDataset, self).__init__()
desired_size = (cfg.TRAIN.SEARCH_SIZE - cfg.TRAIN.EXEMPLAR_SIZE) / \
cfg.ANCHOR.STRIDE + 1 + cfg.TRAIN.BASE_SIZE
if desired_size != cfg.TRAIN.OUTPUT_SIZE:
raise Exception('size not match!')
# Adapt base dataset to tracking dataset loader
sub_dataset = loader(name=name, dataset_root=root)
tracking_sub_dataset = TrackingDatasetAdapter(sub_dataset,
frame_root=frame_root)
self.dataset = tracking_sub_dataset
self.anchor_target = AnchorTarget()
self.template_aug = Augmentation(cfg.DATASET.TEMPLATE.SHIFT,
cfg.DATASET.TEMPLATE.SCALE,
cfg.DATASET.TEMPLATE.BLUR,
cfg.DATASET.TEMPLATE.FLIP,
cfg.DATASET.TEMPLATE.COLOR)
self.search_aug = Augmentation(cfg.DATASET.SEARCH.SHIFT,
cfg.DATASET.SEARCH.SCALE,
cfg.DATASET.SEARCH.BLUR,
cfg.DATASET.SEARCH.FLIP,
cfg.DATASET.SEARCH.COLOR)
def __init__(self,
name,
root,
video_dir,
init_rect,
img_names,
gt_rect,
attr,
load_img=False,
config=None):
self.name = name
self.video_dir = video_dir
self.init_rect = init_rect
self.gt_traj = gt_rect
self.attr = attr
self.pred_trajs = {}
dataset = root.split('/')[-1]
if dataset == 'VOT2018':
self.img_names = [
os.path.join(root, x.replace('color/', '')) for x in img_names
]
elif dataset == 'OTB100':
if video_dir == 'Jogging-1' or 'Skating2-1' in video_dir:
self.img_names = [
os.path.join(root, x.replace('-1', '')) for x in img_names
]
elif video_dir == 'Jogging-2' or 'Skating2-2' in video_dir:
self.img_names = [
os.path.join(root, x.replace('-2', '')) for x in img_names
]
else:
self.img_names = [os.path.join(root, x) for x in img_names]
self.imgs = None
self.config = config
self.size = None
self.center_pos = None
self.template_aug = Augmentation(config.DATASET.TEMPLATE.SHIFT,
config.DATASET.TEMPLATE.SCALE,
config.DATASET.TEMPLATE.BLUR,
config.DATASET.TEMPLATE.FLIP,
config.DATASET.TEMPLATE.COLOR)
self.search_aug = Augmentation(config.DATASET.SEARCH.SHIFT,
config.DATASET.SEARCH.SCALE,
config.DATASET.SEARCH.BLUR,
config.DATASET.SEARCH.FLIP,
config.DATASET.SEARCH.COLOR)
# create anchor target
self.anchor_target = AnchorTarget()
if load_img:
self.imgs = [cv2.imread(x) for x in self.img_names]
self.width = self.imgs[0].shape[1]
self.height = self.imgs[0].shape[0]
else:
img = cv2.imread(self.img_names[0])
assert img is not None, self.img_names[0]
self.width = img.shape[1]
self.height = img.shape[0]