def main():
means = (104, 117, 123) # only support voc now
exp_name = '{}-SSD-{}-{}-bs-{}-{}-lr-{:05d}'.format(
args.net_type, args.dataset, args.input_type, args.batch_size,
args.cfg['base'], int(args.lr * 100000))
args.save_root += args.dataset + '/'
args.listid = '01' ## would be usefull in JHMDB-21
print('Exp name', exp_name, args.listid)
for iteration in [
int(itr) for itr in args.eval_iter.split(',') if len(itr) > 0
]:
log_file = open(
args.save_root + 'cache/' + exp_name +
"/testing-{:d}.log".format(iteration), "w", 1)
log_file.write(exp_name + '\n')
trained_model_path = args.save_root + 'cache/' + exp_name + '/ssd300_ucf24_' + repr(
iteration) + '.pth'
log_file.write(trained_model_path + '\n')
num_classes = len(CLASSES) + 1 #7 +1 background
if args.cfg['base'] == 'fpn':
assert (args.ssd_dim == 512)
net = FPNSSD512(num_classes, args.cfg)
elif args.cfg['base'] == 'vgg16':
if args.cfg['min_dim'][0] == 512:
net = SSD512(num_classes, args.cfg)
net.loc_layers.apply(weights_init)
net.cls_layers.apply(weights_init)
else:
net = build_vgg_ssd(args.num_classes, args.cfg)
net.loc.apply(weights_init)
net.conf.apply(weights_init)
elif args.cfg['base'] == 'mobile_v1_512':
net = MobileSSD512(num_classes, args.cfg)
elif args.cfg['base'] == 'fpn_mobile_512':
net = MobileFPNSSD512(num_classes, args.cfg)
elif args.cfg['base'] == 'yolov3':
net = Darknet('model/yolov3/cfg/yolov3-spp.cfg', arc='CE')
else:
return
net.load_state_dict(torch.load(trained_model_path))
net.eval()
if args.cuda:
net = net.cuda()
cudnn.benchmark = True
print('Finished loading model %d !' % iteration)
# evaluation
torch.cuda.synchronize()
log_file.write('Testing net \n')
test_net(net, args.save_root, exp_name, args.input_type, iteration,
num_classes)
def main():
all_versions = [v1, v2, v3, v4, v5]
args.cfg = all_versions[int(args.version) - 1]
args.basenet = args.cfg['base'] + '_reducedfc.pth'
args.outshape = args.cfg['min_dim']
args.train_sets = 'train'
args.means = (104, 117, 123)
CLASSES = UCF24_CLASSES if args.dataset == 'ucf24' else JHMDB_CLASSES
num_classes = len(CLASSES) + 1
args.num_classes = num_classes
args.stepvalues = [int(val) for val in args.stepvalues.split(',')]
args.loss_reset_step = 30
args.eval_step = 10000
args.print_step = 10
## Define the experiment Name will used to same directory and ENV for visdom
args.exp_name = '{}-SSD-{}-{}-bs-{}-{}-lr-{:05d}'.format(
args.net_type, args.dataset, args.input_type, args.batch_size,
args.cfg['base'], int(args.lr * 100000))
args.save_root += args.dataset + '/'
args.save_root = args.save_root + 'cache/' + args.exp_name + '/'
if not os.path.isdir(args.save_root):
os.makedirs(args.save_root)
def xavier(param):
init.xavier_uniform(param)
def weights_init(m):
if isinstance(m, nn.Conv2d):
xavier(m.weight.data)
m.bias.data.zero_()
if args.cfg['base'] == 'fpn':
net = FPNSSD512(args.num_classes, args.cfg)
elif args.cfg['base'] == 'fpn_cube':
net = FPNSSD512CUBE(args.num_classes, args.cfg)
elif args.cfg['base'] == 'vgg16':
if args.cfg['min_dim'][0] == 512:
net = SSD512(args.num_classes, args.cfg)
net.loc_layers.apply(weights_init)
net.cls_layers.apply(weights_init)
else:
net = build_vgg_ssd(args.num_classes, args.cfg)
net.loc.apply(weights_init)
net.conf.apply(weights_init)
elif args.cfg['base'] == 'fpn_mobile':
net = MobileFPNSSD512(args.num_classes, args.cfg)
elif args.cfg['base'] == 'yolov3':
net = Darknet('model/yolov3/cfg/yolov3-spp.cfg', arc='CE')
else:
return
if args.input_type == 'fastOF':
print(
'Download pretrained brox flow trained model weights and place them at:::=> ',
args.data_root + 'ucf24/train_data/brox_wieghts.pth')
pretrained_weights = args.data_root + 'ucf24/train_data/brox_wieghts.pth'
print('Loading base network...')
net.fpn.load_state_dict(torch.load(pretrained_weights))
elif args.cfg['base'] == 'vgg16' or args.cfg['base'] == 'fpn':
vgg_weights = torch.load(args.data_root + 'ucf24/train_data/' +
args.basenet)
print('Loading base network...')
net.load_weights(vgg_weights)
if args.net_type != 'conv2d' and (args.cfg['base'] == 'vgg16'
or args.cfg['base'] == 'fpn'):
net.transform(args.net_type)
if args.cuda:
net = net.cuda()
args.data_root += args.dataset + '/'
parameter_dict = dict(net.named_parameters(
)) # Get parmeter of network in dictionary format wtih name being key
params = []
#Set different learning rate to bias layers and set their weight_decay to 0
for name, param in parameter_dict.items():
if name.find('bias') > -1:
# print(name, 'layer parameters will be trained @ {}'.format(args.lr*2))
params += [{
'params': [param],
'lr': args.lr * 2,
'weight_decay': 0
}]
else:
# print(name, 'layer parameters will be trained @ {}'.format(args.lr))
params += [{
'params': [param],
'lr': args.lr,
'weight_decay': args.weight_decay
}]
optimizer = optim.SGD(params,
lr=args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
criterion = SphMultiBoxLoss(args.num_classes, 0.5, True, 0, True, 3, 0.5,
False, args.cfg, args.cuda)
scheduler = MultiStepLR(optimizer,
milestones=args.stepvalues,
gamma=args.gamma)
train(args, net, optimizer, criterion, scheduler)
def main():
args.means = (104, 117, 123) # only support voc now
exp_name = '{}-SSD-{}-{}-bs-{}-{}-lr-{:05d}'.format(
args.net_type, args.dataset, args.input_type, args.batch_size,
args.cfg['base'], int(args.lr * 100000))
args.save_root += args.dataset + '/'
args.data_root += args.dataset + '/'
args.listid = '01' ## would be usefull in JHMDB-21
print('Exp name', exp_name, args.listid)
#
for iteration in [
int(itr) for itr in args.eval_iter.split(',') if len(itr) > 0
]:
log_file = open(
args.save_root + 'cache/' + exp_name +
"/testing-{:d}-{:0.2f}.log".format(iteration, args.iou_thresh),
"w", 1)
log_file.write(exp_name + '\n')
trained_model_path = args.save_root + 'cache/' + exp_name + '/ssd300_ucf24_' + repr(
iteration) + '.pth'
log_file.write(trained_model_path + '\n')
num_classes = len(CLASSES) + 1 #7 +1 background
if args.cfg['base'] == 'fpn':
net = FPNSSD512(num_classes, args.cfg)
else:
if args.cfg['min_dim'][0] == 512:
net = SSD512(num_classes, args.cfg)
else:
net = build_vgg_ssd(num_classes, args.cfg)
net.load_state_dict(torch.load(trained_model_path))
net.eval()
if args.cuda:
net = net.cuda()
cudnn.benchmark = True
print('Finished loading model %d !' % iteration)
# Load dataset
if args.dataset == 'ucf24':
dataset = OmniUCF24(args.data_root,
'test',
BaseTransform(300, args.means),
AnnotationTransform(),
input_type=args.input_type,
outshape=args.outshape,
full_test=True)
else:
dataset = OmniJHMDB(args.data_root,
'test',
BaseTransform(300, None),
AnnotationTransform(),
outshape=args.outshape)
# evaluation
torch.cuda.synchronize()
tt0 = time.perf_counter()
log_file.write('Testing net \n')
test_net(net, args.save_root, exp_name, args.input_type, dataset,
iteration, num_classes, log_file)
log_file.close()