def main(train_loader, test_loaders, model, logger, file_logger):
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(args)))
# if (args.enable_logging):
# file_logger.log_string('logs.txt', '\nparsed options:\n{}\n'.format(vars(args)))
if args.cuda:
model.cuda()
optimizer1 = create_optimizer(model.features, args.lr)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print('=> loading checkpoint {}'.format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
else:
print('=> no checkpoint found at {}'.format(args.resume))
start = args.start_epoch
end = start + args.epochs
for epoch in range(start, end):
# iterate over test loaders and test results
#train_loader, test_loaders2 = create_loaders(load_random_triplets=triplet_flag)
train(train_loader, model, optimizer1, epoch, logger, triplet_flag)
for test_loader in test_loaders:
test(test_loader['dataloader'], model, epoch, logger, test_loader['name'])
if TEST_ON_W1BS:
# print(weights_path)
patch_images = w1bs.get_list_of_patch_images(
DATASET_DIR=args.w1bsroot.replace('/code', '/data/W1BS'))
desc_name = 'curr_desc' # + str(random.randint(0,100))
DESCS_DIR = LOG_DIR + '/temp_descs/' # args.w1bsroot.replace('/code', "/data/out_descriptors")
OUT_DIR = DESCS_DIR.replace('/temp_descs/', "/out_graphs/")
for img_fname in patch_images:
w1bs_extract_descs_and_save(img_fname, model, desc_name, cuda=args.cuda,
mean_img=args.mean_image,
std_img=args.std_image, out_dir=DESCS_DIR)
force_rewrite_list = [desc_name]
w1bs.match_descriptors_and_save_results(DESC_DIR=DESCS_DIR, do_rewrite=True,
dist_dict={},
force_rewrite_list=force_rewrite_list)
if (args.enable_logging):
w1bs.draw_and_save_plots_with_loggers(DESC_DIR=DESCS_DIR, OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name],
logger=file_logger,
tensor_logger=logger)
else:
w1bs.draw_and_save_plots(DESC_DIR=DESCS_DIR, OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name])
def main(train_loader, test_loaders, model, logger, file_logger):
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(args)))
#if (args.enable_logging):
# file_logger.log_string('logs.txt', '\nparsed options:\n{}\n'.format(vars(args)))
if args.cuda:
model.cuda()
# change model.features to model.parameters()
optimizer1 = create_optimizer(model, args.lr)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print('=> loading checkpoint {}'.format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
try:
model.load_state_dict(checkpoint['state_dict'])
except:
t=0
for i in model.state_dict():
if(i.startswith('features') or i.startswith('feature_L2NET')):
model.state_dict()[i]=checkpoint['state_dict'].items()[t]
t=t+1
print(i+" is loaded")
args.start_epoch=0
try:
optimizer1.load_state_dict(checkpoint['optimizer'])
print('=> optimizer loaded from {}'.format(args.resume))
print('current lr is {}'.format(optimizer1.param_groups[0]['lr']))
for group in optimizer1.param_groups:
if 'step' not in group:
group['step'] = 0.
else:
print('current step is {}'.format(group['step']))
# group['step'] = (args.n_triplets)*args.start_epoch//args.batch_size+1
except:
print('=> on optimizer saved in {}'.format(args.resume))
else:
print('=> no checkpoint found at {}'.format(args.resume))
start = args.start_epoch
end = args.end_epoch if args.end_epoch < (start + args.epochs) else (start + args.epochs)
trainloss=[]
testacc=[]
for epoch in range(start, end):
# iterate over test loaders and test results
train(test_loaders[0], train_loader, model, optimizer1, epoch, logger, triplet_flag)
for test_loader in test_loaders:
test(test_loader['dataloader'], model, epoch, logger, test_loader['name'])
if args.testbinary:
test_binary(test_loader['dataloader'], model, epoch, logger, test_loader['name'])
x = copy.copy(dataset_names)
x.remove(args.training_set)
old_trainingset=args.training_set
args.training_set = x[0]
args.training_set = old_trainingset
if TEST_ON_W1BS :
# print(weights_path)
patch_images = w1bs.get_list_of_patch_images(
DATASET_DIR=args.w1bsroot.replace('/code', '/data/W1BS'))
desc_name = 'curr_desc'# + str(random.randint(0,100))
DESCS_DIR = LOG_DIR + '/temp_descs/' #args.w1bsroot.replace('/code', "/data/out_descriptors")
OUT_DIR = DESCS_DIR.replace('/temp_descs/', "/out_graphs/")
for img_fname in patch_images:
w1bs_extract_descs_and_save(img_fname, model, desc_name, cuda = args.cuda,
mean_img=args.mean_image,
std_img=args.std_image, out_dir = DESCS_DIR)
force_rewrite_list = [desc_name]
w1bs.match_descriptors_and_save_results(DESC_DIR=DESCS_DIR, do_rewrite=True,
dist_dict={},
force_rewrite_list=force_rewrite_list)
train_loader, test_loaders2 = create_loaders(load_random_triplets=triplet_flag)
def main(trainPhotoTourDataset, test_loaders, model, logger, file_logger):
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(args)))
if (args.enable_logging):
file_logger.log_string('logs.txt', '\nparsed options:\n{}\n'.format(vars(args)))
if args.cuda:
model.cuda()
optimizer1 = create_optimizer(model.features, args)
# optionally resume from a checkpoint
if args.resume:
model = resume_pretrained_model(model, args)
start = args.start_epoch
end = start + args.epochs
kwargs = {'num_workers': args.num_workers, 'pin_memory': args.pin_memory} if args.cuda else {}
transform = transforms.Compose([
transforms.Lambda(cv2_scale),
transforms.Lambda(np_reshape),
transforms.ToTensor(),
transforms.Normalize((args.mean_image,), (args.std_image,))])
for epoch in xrange(start, end):
model.eval()
# #
descriptors = get_descriptors_for_dataset(model, trainPhotoTourDataset)
#
np.save('descriptors.npy', descriptors)
descriptors = np.load('descriptors.npy')
#
hard_negatives = get_hard_negatives(trainPhotoTourDataset, descriptors)
np.save('descriptors_min_dist.npy', hard_negatives)
hard_negatives = np.load('descriptors_min_dist.npy')
print(hard_negatives[0])
trainPhotoTourDatasetWithHardNegatives = TripletPhotoTourHardNegatives(train=True,
negative_indices=hard_negatives,
batch_size=args.batch_size,
root=args.dataroot,
name=args.training_set,
download=True,
transform=transform)
train_loader = torch.utils.data.DataLoader(trainPhotoTourDatasetWithHardNegatives,
batch_size=args.batch_size,
shuffle=False, **kwargs)
train(train_loader, model, optimizer1, epoch, logger)
# iterate over test loaders and test results
for test_loader in test_loaders:
test(test_loader['dataloader'], model, epoch, logger, test_loader['name'])
if TEST_ON_W1BS :
# print(weights_path)
patch_images = w1bs.get_list_of_patch_images(
DATASET_DIR=args.w1bsroot.replace('/code', '/data/W1BS'))
desc_name = 'curr_desc'
for img_fname in patch_images:
w1bs_extract_descs_and_save(img_fname, model, desc_name, cuda = args.cuda,
mean_img=args.mean_image,
std_img=args.std_image)
DESCS_DIR = args.w1bsroot.replace('/code', "/data/out_descriptors")
OUT_DIR = args.w1bsroot.replace('/code', "/data/out_graphs")
force_rewrite_list = [desc_name]
w1bs.match_descriptors_and_save_results(DESC_DIR=DESCS_DIR, do_rewrite=True,
dist_dict={},
force_rewrite_list=force_rewrite_list)
if(args.enable_logging):
w1bs.draw_and_save_plots_with_loggers(DESC_DIR=DESCS_DIR, OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name],
logger=file_logger,
tensor_logger = None)
else:
w1bs.draw_and_save_plots(DESC_DIR=DESCS_DIR, OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name],
really_draw = False)
def execute(self, train_loader, test_loaders, model, logger, file_logger):
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(self.args)))
#if (self.args.enable_logging):
# file_logger.log_string('logs.txt', '\nparsed options:\n{}\n'.format(vars(self.args)))
if self.args.cuda:
model.cuda()
optimizer1 = self.create_optimizer(model.features, self.args.lr)
# optionally resume from a checkpoint
if self.args.resume:
if os.path.isfile(self.args.resume):
print('=> loading checkpoint {}'.format(self.args.resume))
checkpoint = torch.load(self.args.resume)
self.args.start_epoch = checkpoint['epoch']
checkpoint = torch.load(self.args.resume)
model.load_state_dict(checkpoint['state_dict'])
else:
print('=> no checkpoint found at {}'.format(self.args.resume))
start = self.args.start_epoch
end = start + self.args.epochs
for epoch in range(start, end):
# iterate over test loaders and test results
self.train(train_loader, model, optimizer1, epoch, logger, self.triplet_flag)
for test_loader in test_loaders:
self.test(test_loader['dataloader'], model, epoch, logger, test_loader['name'])
if self.test_on_w1bs:
print("Saving test_on_w1bs results")
patch_images = w1bs.get_list_of_patch_images(
DATASET_DIR=self.args.w1bsroot)
desc_name = 'curr_desc'# + str(random.randint(0,100))
self.descs_dir = self.log_dir + '/temp_descs/' #self.args.w1bsroot.replace('/code', "/data/out_descriptors")
OUT_DIR = self.descs_dir.replace('/temp_descs/', "/out_graphs/")
for img_fname in patch_images:
w1bs_extract_descs_and_save(img_fname, model, desc_name, cuda = self.args.cuda,
mean_img=self.args.mean_image,
std_img=self.args.std_image, out_dir = self.descs_dir)
force_rewrite_list = [desc_name]
w1bs.match_descriptors_and_save_results(DESC_DIR=self.descs_dir, do_rewrite=True,
dist_dict={},
force_rewrite_list=force_rewrite_list)
print("descs_dir", self.descs_dir)
print("OUT_DIR", OUT_DIR)
print("Number of patch_images", len(patch_images))
if(self.args.enable_logging):
w1bs.draw_and_save_plots_with_loggers(DESC_DIR=self.descs_dir, OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name],
logger=file_logger,
tensor_logger = logger)
else:
w1bs.draw_and_save_plots(DESC_DIR=self.descs_dir, OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name])
#randomize train loader batches
train_loader, test_loaders2 = self.create_loaders(load_random_triplets=self.triplet_flag)
def main(train_loader, test_loaders, model, logger, file_logger):
# print the experiment configuration
print('\nparsed options:\n{}\n'.format(vars(args)))
if (args.enable_logging):
file_logger.log_string('logs',
'\nparsed options:\n{}\n'.format(vars(args)))
if args.cuda:
model.cuda()
optimizer1 = create_optimizer(model.features, args)
# optionally resume from a checkpoint
if args.resume:
model = resume_pretrained_model(model, args, suffix)
start = args.start_epoch
end = start + args.epochs
for epoch in xrange(start, end):
# iterate over test loaders and test results
train(train_loader, model, optimizer1, epoch, file_logger,
triplet_flag)
for test_loader in test_loaders:
test(test_loader['dataloader'], model, epoch, file_logger,
test_loader['name'])
if TEST_ON_W1BS:
# print(weights_path)
patch_images = w1bs.get_list_of_patch_images(
DATASET_DIR=args.w1bsroot.replace('/code', '/data/W1BS'))
desc_name = 'curr_desc' # + str(random.randint(0,100))
DESCS_DIR = log_dir + '/temp_descs/' #args.w1bsroot.replace('/hardnet', "/data/out_descriptors")
OUT_DIR = DESCS_DIR.replace('/temp_descs/', "/out_graphs/")
for img_fname in patch_images:
# a bug occur @func
# save (patch*feat_dim) desc matrix to the desc file
w1bs_extract_descs_and_save(img_fname,
model,
desc_name,
cuda=args.cuda,
mean_img=args.mean_image,
std_img=args.std_image,
out_dir=DESCS_DIR)
force_rewrite_list = [desc_name]
w1bs.match_descriptors_and_save_results(
DESC_DIR=DESCS_DIR,
do_rewrite=True,
dist_dict={},
force_rewrite_list=force_rewrite_list)
if (args.enable_logging):
# DESC_DIR, OUT_DIR, methods, colors, lines,
# descs_to_draw, really_draw, logger, tensor_logger = None
w1bs.draw_and_save_plots_with_loggers(
DESC_DIR=DESCS_DIR,
OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name],
really_draw=True,
logger=file_logger,
tensor_logger=logger)
else:
w1bs.draw_and_save_plots(DESC_DIR=DESCS_DIR,
OUT_DIR=OUT_DIR,
methods=["SNN_ratio"],
descs_to_draw=[desc_name],
really_draw=True)
# re-generate the triplets slices
if epoch + 1 < end:
train_loader.resample_dataset_triplets()
print('HardNet train done!')
