def pre_init_with_sift(trainPhotoTourDataset):
patch_size = 65
ON_GPU = True
SIFT = SIFTNet(patch_size=patch_size, do_cuda=ON_GPU)
SIFT.eval()
if ON_GPU:
SIFT.cuda()
transformed = []
for img in trainPhotoTourDataset.data:
transformed.append(
np.expand_dims(cv2.resize(img.cpu().numpy(), (65, 65)), axis=0))
phototour_loader = data_utils.DataLoader(
PhototourTrainingData(transformed), batch_size=256, shuffle=False)
descriptors = []
pbar = tqdm(enumerate(phototour_loader))
for batch_idx, data_a in pbar:
if ON_GPU:
torch_patches = Variable(data_a.type(torch.FloatTensor).cuda(),
volatile=True)
else:
torch_patches = Variable(data_a.type(torch.FloatTensor),
volatile=True)
res = SIFT(torch_patches)
sift = np.round(512. * res.data.cpu().numpy()).astype(np.float32)
descriptors.extend(sift)
return np.array(descriptors)
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
cudnn.benchmark = True
torch.cuda.manual_seed_all(args.seed)
# create loggin directory
if not os.path.exists(args.log_dir):
os.makedirs(args.log_dir)
# set random seeds
torch.manual_seed(args.seed)
np.random.seed(args.seed)
if args.descriptor == 'SIFT':
descriptor = SIFTNet(patch_size=32)
if not args.no_cuda:
descriptor = descriptor.cuda()
elif args.descriptor == 'HardNet':
descriptor = HardNet()
#descriptor = HardNetNarELU(SIFTNet(patch_size=32))
if not args.no_cuda:
descriptor = descriptor.cuda()
model_weights = 'HardNet++.pth'
#model_weights = 'HardNetELU_Narr.pth'
hncheckpoint = torch.load(model_weights)
descriptor.load_state_dict(hncheckpoint['state_dict'])
descriptor.train()
else:
descriptor = lambda x: x.view(x.size(0),-1)
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
cudnn.benchmark = True
torch.cuda.manual_seed_all(args.seed)
# create loggin directory
if not os.path.exists(args.log_dir):
os.makedirs(args.log_dir)
# set random seeds
torch.manual_seed(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
if args.descriptor == 'SIFT':
descriptor = SIFTNet(patch_size=PS)
if not args.no_cuda:
descriptor = descriptor.cuda()
elif args.descriptor == 'HardNet':
descriptor = HardNet()
if not args.no_cuda:
descriptor = descriptor.cuda()
model_weights = 'HardNet++.pth'
hncheckpoint = torch.load(model_weights)
descriptor.load_state_dict(hncheckpoint['state_dict'])
descriptor.train()
elif args.descriptor == 'TFeat':
descriptor = HardTFeatNet(sm=SIFTNet(patch_size=32))
if not args.no_cuda:
descriptor = descriptor.cuda()
model_weights = 'HardTFeat.pth'
return var_image_reshape
from HardNet import HardNet, HardTFeatNet, L2Norm
from pytorch_sift import SIFTNet
import math
d1 = lambda x: L2Norm()(x.view(x.size(0), -1) - x.view(x.size(0), -1).mean(
dim=1, keepdim=True).expand(x.size(0),
x.size(1) * x.size(2) * x.size(3)).detach())
d2 = HardNet()
model_weights = '../../HardNet++.pth'
hncheckpoint = torch.load(model_weights)
d2.load_state_dict(hncheckpoint['state_dict'])
d2.eval()
d3 = SIFTNet(patch_size=32)
model_weights = 'HardTFeat.pth'
d4 = HardTFeatNet(sm=SIFTNet(patch_size=32))
checkpoint = torch.load(model_weights)
d4.load_state_dict(checkpoint['state_dict'])
d4 = nn.Sequential(d4, L2Norm())
desc_list = [d1, d2, d3, d4]
desc_names = ['Pixels', 'HardNet', 'SIFT', 'TFeat']
USE_CUDA = False
detector = ScaleSpaceAffinePatchExtractor(mrSize=5.12,
num_features=200,
border=32,
num_Baum_iters=0)
descriptor = HardNet()
5.0, scale_log = 0.4)
if len(geom_loss.size()) == 0:
print 'skip'
continue
print 'A', A1.data.cpu().numpy()[0:1,:,:]
#print fro_dists
loss = geom_loss.mean()
total_feats += geom_loss.size(0)
total_loss += loss.data.cpu().numpy()[0]
print 'test img', batch_idx, loss.data.cpu().numpy()[0], geom_loss.size(0)
print 'Total loss:', total_loss / float(batch_idx+1), 'features', float(total_feats) / float(batch_idx+1)
model.num = model_num_feats
train_loader, test_loader = create_loaders()
HA = ScaleSpaceAffinePatchExtractor( mrSize = 5.192, num_features = 350, border = 5, num_Baum_iters = 1, AffNet = LAFNet())
from pytorch_sift import SIFTNet
SIFT = SIFTNet(patch_size = 32 )
if USE_CUDA:
HA = HA.cuda()
optimizer1 = create_optimizer(HA.AffNet, BASE_LR, 1e-4)
#test(test_loader, model, cuda = USE_CUDA)
for epoch in range(n_epochs):
print 'epoch', epoch
train(train_loader, HA, optimizer1, epoch, cuda = USE_CUDA)
test(test_loader, HA, cuda = USE_CUDA)
for t in self.itr:
im_path = os.path.join(base, t + '.png')
im = cv2.imread(im_path, 0)
self.N = im.shape[0] / 65
setattr(self, t, np.split(im, self.N))
seqs = glob.glob(sys.argv[1] + '/*')
seqs = [os.path.abspath(p) for p in seqs]
descr_name_vl = 'pytorch-sift-vlfeat-' + str(OUT_W)
descr_name_mp = 'pytorch-sift-mp-' + str(OUT_W)
descr_names = [descr_name_vl, descr_name_mp]
model_vl = SIFTNet(OUT_W, mask_type='Gauss', sigma_type='vlfeat')
model_vl.cuda()
print(model_vl)
model_mp = SIFTNet(OUT_W, mask_type='CircularGauss', sigma_type='hesamp')
model_mp.cuda()
print(model_mp)
models = [model_vl, model_mp]
for seq_path in seqs:
seq = hpatches_sequence(seq_path)
for descr_id, descr_name in enumerate(descr_names):
model = models[descr_id]
path = os.path.join(descr_name, seq.name)
if not os.path.exists(path):
min_tilt=1.0,
max_tilt=1.0,
max_rot=1.0,
max_shift=0.5,
mrSize=1.0,
use_cuda=USE_CUDA)
def weights_init(m):
if isinstance(m, nn.Conv2d):
nn.init.xavier_normal(m.weight.data)
#SIRNet = SparseImgRepresenter(detector_net = ConvST_net, descriptor_net = hardnet)
SIRNet = SparseImgRepresenter(detector_net=ConvST_net,
descriptor_net=SIFTNet(patch_size=16,
do_cuda=USE_CUDA))
SIRNet.detector.apply(weights_init)
#aff_norm_patches, LAFs, descs = SIRNet(var_image_reshape)
def distance_matrix_vector(anchor, positive):
"""Given batch of anchor descriptors and positive descriptors calculate distance matrix"""
d1_sq = torch.sum(anchor * anchor, dim=1)
d2_sq = torch.sum(positive * positive, dim=1)
eps = 1e-6
return torch.sqrt(
torch.abs((d1_sq.expand(positive.size(0), anchor.size(0)) +
torch.t(d2_sq.expand(anchor.size(0), positive.size(0))) -
2.0 * torch.bmm(positive.unsqueeze(0),
torch.t(anchor).unsqueeze(0)).squeeze(0)) +
assert ((height % patch_h == 0) and (width % patch_w == 0))
patch_idxs = []
patches = []
current_patch_idx = start_patch_idx
for y in range(0, height, patch_h):
patch_idxs.append([])
for x in range(0, width, patch_w):
patch = np.array(img.crop(
(x, y, x + patch_w, y + patch_h))).astype(np.uint8)
patches.append(patch)
patch_idxs[-1].append(current_patch_idx)
current_patch_idx += 1
return np2torch(np.array(patches)), patch_idxs, patch_idxs[-1][-1]
model = SIFTNet(patch_size=41)
model.cuda()
model.eval()
try:
input_img_fname = sys.argv[1]
output_fname = sys.argv[2]
except:
print "Wrong input format. Try ./rank_scale_patches.py imgs/ref.png out.txt"
sys.exit(1)
patches, idxs, max_idx = read_patch_file(input_img_fname,
patch_w=73,
patch_h=41)
bw_patches = patches.float().mean(dim=1, keepdim=True)
print bw_patches.shape
