def main():
args = parse_args()
# prepare model (SiamRPN or SiamFC)
# prepare tracker
info = edict()
info.arch = args.arch
info.dataset = args.video
info.epoch_test = True
info.cls_type = 'thinner'
if 'FC' in args.arch:
net = models.__dict__[args.arch]()
tracker = SiamFC(info)
else:
net = models.__dict__[args.arch](anchors_nums=5, cls_type='thinner')
tracker = SiamRPN(info)
print('[*] ======= Track video with {} ======='.format(args.arch))
net = load_pretrain(net, args.resume)
net.eval()
net = net.cuda()
# check init box is list or not
if not isinstance(args.init_bbox, list) and args.init_bbox is not None:
args.init_bbox = list(eval(args.init_bbox))
else:
pass
track_video(tracker, net, args.video, init_box=args.init_bbox)
def main():
args = parse_args()
# prepare model (SiamRPN or SiamFC)
# prepare tracker
info = edict()
info.arch = args.arch
info.dataset = 'NOUSE'
info.epoch_test = True
if 'FC' in args.arch:
net = models.__dict__[args.arch]()
tracker = SiamFC(info)
else:
net = models.__dict__[args.arch](anchors_nums=5)
tracker = SiamRPN(info)
print('[*] ======= Track video with {} ======='.format(args.arch))
net = load_pretrain(net, args.resume)
net.eval()
net = net.cuda()
track_webcam(tracker, net)
def __init__(self, backbone, target):
super(SiamDWGroup, self).__init__(f"SiamDWGroup/{backbone}/{target}")
info = edict()
info.arch = backbone
info.dataset = target
info.epoch_test = False
if target == "OTB":
info.dataset = "OTB2015"
elif target == "VOT":
info.dataset = "VOT2017"
else:
raise ValueError("Invalid target")
if backbone == "SiamFCIncep22":
net_file = path_config.SIAMDW_CIRINCEP22_MODEL
self.tracker = SiamFC(info)
self.net = models.__dict__[info.arch]()
elif backbone == "SiamFCNext22":
net_file = path_config.SIAMDW_CIRNEXT22_MODEL
self.tracker = SiamFC(info)
self.net = models.__dict__[info.arch]()
elif backbone == "SiamFCRes22":
net_file = path_config.SIAMDW_CIRESNET22_MODEL
self.tracker = SiamFC(info)
self.net = models.__dict__[info.arch]()
elif backbone == "SiamRPNRes22":
net_file = path_config.SIAMDW_CIRESNET22_RPN_MODEL
info.cls_type = "thinner"
self.tracker = SiamRPN(info)
self.net = models.__dict__[info.arch](anchors_nums=5,
cls_type=info.cls_type)
else:
raise ValueError("Invalid backbone")
self.net = load_pretrain(self.net, net_file)
self.net.eval()
self.net = self.net.cuda()
def main():
args = parse_args()
# prepare model
net = models.__dict__[args.arch]()
net = load_pretrain(net, args.resume)
net.eval()
net = net.cuda()
# prepare video
dataset = load_dataset(args.dataset)
video_keys = list(dataset.keys()).copy()
# prepare tracker
info = edict()
info.arch = args.arch
info.dataset = args.dataset
info.epoch_test = args.epoch_test
tracker = SiamFC(info)
# tracking all videos in benchmark
for video in video_keys:
track(tracker, net, dataset[video], args)
crossover=crossover, mutation=mutation, \
analysis=[FitnessStore, ConsoleOutput])
# create model
net = models.__dict__[args.arch]()
net = load_pretrain(net, args.resume)
net.eval()
net = net.cuda()
print('==> pretrained model has been loaded')
# prepare tracker
info = edict()
info.arch = args.arch
info.dataset = args.dataset
info.epoch_test = False
tracker = SiamFC(info)
# Define fitness function.
@engine.fitness_register
def fitness(indv):
scale_step, scale_lr, scale_penalty, window_influence = indv.solution
# add params top config
config = dict()
config['benchmark'] = args.dataset
config['arch'] = args.arch
config['resume'] = args.resume
config['hp'] = dict()
config['hp']['scale_step'] = scale_step
config['hp']['scale_penalty'] = scale_penalty
def fitness(config, reporter):
# create model
if args.arch in ['Ocean']:
model = models.__dict__[args.arch](align=info.align)
tracker = Ocean(info)
elif args.arch in ['OceanPlus']:
model = models.__dict__[args.arch](online=args.online, mms=args.MMS)
tracker = OceanPlus(info)
elif 'SiamDW' in args.arch:
model = models.__dict__[args.arch]()
tracker = SiamFC(info)
else:
raise ValueError('not supported other model now')
model = load_pretrain(model, args.resume)
model.eval()
model = model.cuda()
print('pretrained model has been loaded')
print(os.environ['CUDA_VISIBLE_DEVICES'])
if args.arch in ['Ocean']:
penalty_k = config["penalty_k"]
scale_lr = config["scale_lr"]
window_influence = config["window_influence"]
small_sz = config["small_sz"]
big_sz = config["big_sz"]
ratio = config["ratio"]
model_config = dict()
model_config['benchmark'] = args.dataset
model_config['arch'] = args.arch
model_config['resume'] = args.resume
model_config['hp'] = dict()
model_config['hp']['penalty_k'] = penalty_k
model_config['hp']['window_influence'] = window_influence
model_config['hp']['lr'] = scale_lr
model_config['hp']['small_sz'] = small_sz
model_config['hp']['big_sz'] = big_sz
model_config['hp']['ratio'] = ratio
elif args.arch in ['OceanPlus']:
penalty_k = config["penalty_k"]
scale_lr = config["scale_lr"]
window_influence = config["window_influence"]
small_sz = config["small_sz"]
big_sz = config["big_sz"]
lambda_u = config["lambda_u"]
lambda_s = config["lambda_s"]
cyclic_thr = config["cyclic_thr"]
choose_thr = config["choose_thr"]
model_config = dict()
model_config['benchmark'] = args.dataset
model_config['arch'] = args.arch
model_config['resume'] = args.resume
model_config['hp'] = dict()
model_config['hp']['penalty_k'] = penalty_k
model_config['hp']['window_influence'] = window_influence
model_config['hp']['lr'] = scale_lr
model_config['hp']['small_sz'] = small_sz
model_config['hp']['big_sz'] = big_sz
model_config['hp']['lambda_u'] = lambda_u
model_config['hp']['lambda_s'] = lambda_s
model_config['hp']['cyclic_thr'] = cyclic_thr
model_config['hp']['choose_thr'] = choose_thr
elif args.arch in ['SiamFC', 'SiamDW']:
scale_step = config["scale_step"]
scale_penalty = config["scale_penalty"]
scale_lr = config["scale_lr"]
w_influence = config["w_influence"]
model_config = dict()
model_config['benchmark'] = args.dataset
model_config['arch'] = args.arch
model_config['resume'] = args.resume
model_config['hp'] = dict()
model_config['hp']['scale_step'] = scale_step
model_config['hp']['scale_penalty'] = scale_penalty
model_config['hp']['w_influence'] = w_influence
model_config['hp']['scale_lr'] = scale_lr
else:
raise ValueError('not supported tracker')
# -------------
# For OceanPlus
# -------------
# VOT and OceanPlus
if args.dataset.startswith('VOT') and args.arch in ['OceanPlus']:
eao = eao_vot_oceanplus(tracker, model, model_config)
print(
"penalty_k: {0}, scale_lr: {1}, window_influence: {2}, small_sz: {3}, big_sz: {4}, lambda_u: {5}, lambda_s: {6}, cyclic_thr: {7}, choose_thr: {8}, eao: {9}"
.format(penalty_k, scale_lr, window_influence, small_sz, big_sz,
lambda_u, lambda_s, cyclic_thr, choose_thr, eao))
reporter(EAO=eao)
# GOT10K and OceanPlus
if args.dataset.startswith('GOT') and args.arch in ['OceanPlus']:
auc = auc_got10k_oceanplus(tracker, model, model_config)
print(
"penalty_k: {0}, scale_lr: {1}, window_influence: {2}, small_sz: {3}, big_sz: {4}, lambda_u: {5}, lambda_s: {6}, cyclic_thr: {7}, choose_thr: {8}, eao: {9}"
.format(penalty_k, scale_lr, window_influence, small_sz, big_sz,
lambda_u, lambda_s, cyclic_thr, choose_thr, auc.item()))
reporter(AUC=auc)
# ----------
# For Ocean
# ----------
# VOT and Ocean
if args.dataset.startswith('VOT') and args.arch in ['Ocean']:
eao = eao_vot(tracker, model, model_config)
print(
"penalty_k: {0}, scale_lr: {1}, window_influence: {2}, small_sz: {3}, big_sz: {4}, ratio: {6}, eao: {5}"
.format(penalty_k, scale_lr, window_influence, small_sz, big_sz,
eao, ratio))
reporter(EAO=eao)
# OTB and Ocean
if args.dataset.startswith('OTB') and args.arch in ['Ocean']:
auc = auc_otb(tracker, model, model_config)
print(
"penalty_k: {0}, scale_lr: {1}, window_influence: {2}, small_sz: {3}, big_sz: {4}, ratio: {6}, eao: {5}"
.format(penalty_k, scale_lr, window_influence, small_sz, big_sz,
auc.item(), ratio))
reporter(AUC=auc)
# ------------
# For SiamDW
# ------------
# OTB and SiamFC/DW
if args.dataset.startswith('OTB') and ('SiamFC' in args.arch
or 'SiamDW' in args.arch):
auc = auc_otb_fc(tracker, model, model_config)
print(
"scale_step: {0}, scale_lr: {1}, scale_penalty: {2}, window_influence: {3}, auc: {4}"
.format(scale_step, scale_lr, scale_penalty, w_influence,
auc.item()))
reporter(AUC=auc)
# VOT and SiamFC/DW
if args.dataset.startswith('VOT') and ('SiamFC' in args.arch
or 'SiamDW' in args.arch):
eao = eao_vot_fc(tracker, model, model_config)
print(
"scale_step: {0}, scale_lr: {1}, scale_penalty: {2}, window_influence: {3}, eao: {4}"
.format(scale_step, scale_lr, scale_penalty, w_influence, eao))
reporter(EAO=eao)
