img = np.mean(np.array(img), axis=2)
var_image = torch.autograd.Variable(torch.from_numpy(img.astype(np.float32)),
volatile=True)
var_image_reshape = var_image.view(1, 1, var_image.size(0), var_image.size(1))
AffNetPix = AffNetFast(PS=32)
weightd_fname = '../../pretrained/AffNet.pth'
checkpoint = torch.load(weightd_fname)
AffNetPix.load_state_dict(checkpoint['state_dict'])
AffNetPix.eval()
HA = ScaleSpaceAffinePatchExtractor(mrSize=5.192,
num_features=nfeats,
border=5,
num_Baum_iters=1,
th=th,
AffNet=AffNetPix)
if USE_CUDA:
HA = HA.cuda()
var_image_reshape = var_image_reshape.cuda()
with torch.no_grad():
LAFs, resp = HA(var_image_reshape)
ells = LAFs2ell(LAFs.data.cpu().numpy())
np.savetxt(output_fname, ells, delimiter=' ', fmt='%10.10f')
line_prepender(output_fname, str(len(ells)))
line_prepender(output_fname, '1.0')
final_aff_m,
final_pyr_idxs,
final_level_idxs,
PS=19,
pyr_inv_idxs=pyr_inv_idxs)
###
patches = self.extract_patches(scale_pyr,
final_aff_m,
final_pyr_idxs,
final_level_idxs,
PS=self.PS,
pyr_inv_idxs=pyr_inv_idxs)
return final_aff_m, patches, final_resp, scale_pyr
HA = HessianAffinePatchExtractor(mrSize=5.0,
num_features=3000,
border=1,
num_Baum_iters=5)
aff, patches, resp, pyr = HA(var_image_reshape / 255.)
LAFs = aff.data.cpu().numpy()
LAFs[:, 0, 2] *= float(img.shape[1])
LAFs[:, 1, 2] *= float(img.shape[0])
LAFs[:, :2, :2] *= min(float(img.shape[0]), float(img.shape[0]))
ells = LAFs2ell(LAFs)
np.savetxt(output_fname, ells, delimiter=' ', fmt='%10.10f')
#line_prepender(output_fname, str(len(ells)))
#line_prepender(output_fname, '1.0')
descriptors_for_net = np.zeros((n_patches, 128))
bs = 128;
outs = []
n_batches = n_patches / bs + 1
t = time.time()
patches = patches.cuda()
hardnet = hardnet.cuda()
for batch_idx in range(n_batches):
if batch_idx == n_batches - 1:
if (batch_idx + 1) * bs > n_patches:
end = n_patches
else:
end = (batch_idx + 1) * bs
else:
end = (batch_idx + 1) * bs
data_a = patches[batch_idx * bs: end, :, :, :]
out_a = hardnet(data_a)
descriptors_for_net[batch_idx * bs: end,:] = out_a.data.cpu().numpy().reshape(-1, 128)
print descriptors_for_net.shape
et = time.time() - t
print 'processing', et, et/float(n_patches), ' per patch'
ells = convert_LAF_to_this_stupid_Oxford_ellipse_format(var_image_reshape.data.cpu().numpy()[0,0,:,:], LAFs)
np.savetxt(output_fname, np.hstack([ells, descriptors_for_net]), delimiter=' ', fmt='%10.9f')
'''
ells = LAFs2ell(var_image_reshape.data.cpu().numpy()[0, 0, :, :], LAFs)
np.savetxt(output_fname, ells, delimiter=' ', fmt='%10.10f')
line_prepender(output_fname, str(len(ells)))
line_prepender(output_fname, '1.0')