def save_network(self, dcn, optimizer, iteration, logging_dict=None):
"""
Saves network parameters to logging directory
:return:
:rtype: None
"""
network_param_file = os.path.join(
self._logging_dir,
utils.getPaddedString(iteration, width=6) + ".pth")
optimizer_param_file = network_param_file + ".opt"
torch.save(dcn.state_dict(), network_param_file)
torch.save(optimizer.state_dict(), optimizer_param_file)
# also save loss history stuff
if logging_dict is not None:
log_history_file = os.path.join(
self._logging_dir,
utils.getPaddedString(iteration, width=6) +
"_log_history.yaml")
utils.saveToYaml(logging_dict, log_history_file)
current_loss_file = os.path.join(self._logging_dir, 'loss.yaml')
current_loss_data = self._get_current_loss(logging_dict)
utils.saveToYaml(current_loss_data, current_loss_file)
def find_best_match(self, event, u, v, flags, param):
"""
For each network, find the best match in the target image to point highlighted
with reticle in the source image. Displays the result
:return:
:rtype:
"""
img_1_with_reticle = np.copy(self.img1)
draw_reticle(img_1_with_reticle, u, v, self._reticle_color)
cv2.imshow("source", img_1_with_reticle)
alpha = self._config["blend_weight_original_image"]
beta = 1 - alpha
img_2_with_reticle = np.copy(self.img2)
print "\n\n"
self._res_uv = dict()
# self._res_a_uv = dict()
# self._res_b_uv = dict()
for network_name in self._dcn_dict:
res_a = self._res_a[network_name]
res_b = self._res_b[network_name]
best_match_uv, best_match_diff, norm_diffs = \
DenseCorrespondenceNetwork.find_best_match((u, v), res_a, res_b)
print "\n\n"
print "network_name:", network_name
self._res_uv[network_name] = dict()
self._res_uv[network_name]['source'] = res_a[v, u, :].tolist()
self._res_uv[network_name]['target'] = res_b[v, u, :].tolist()
# print "res_a[v, u, :]:", res_a[v, u, :]
# print "res_b[v, u, :]:", res_b[v, u, :]
print "%s best match diff: %.3f" % (network_name, best_match_diff)
threshold = self._config["norm_diff_threshold"]
if network_name in self._config["norm_diff_threshold_dict"]:
threshold = self._config["norm_diff_threshold_dict"][
network_name]
heatmap = self.scale_norm_diffs_to_make_heatmap(
norm_diffs, threshold)
reticle_color = self._network_reticle_color[network_name]
draw_reticle(heatmap, best_match_uv[0], best_match_uv[1],
reticle_color)
draw_reticle(img_2_with_reticle, best_match_uv[0],
best_match_uv[1], reticle_color)
blended = cv2.addWeighted(self.img2_gray, alpha, heatmap, beta, 0)
cv2.imshow(network_name, blended)
cv2.imshow("target", img_2_with_reticle)
if event == cv2.EVENT_LBUTTONDOWN:
utils.saveToYaml(self._res_uv, 'clicked_point.yaml')
def find_best_match(self, event,u,v,flags,param):
"""
For each network, find the best match in the target image to point highlighted
with reticle in the source image. Displays the result
:return:
:rtype:
"""
img_1_with_reticle = np.copy(self.img1)
draw_reticle(img_1_with_reticle, u, v, self._reticle_color)
cv2.imshow("source", img_1_with_reticle)
alpha = self._config["blend_weight_original_image"]
beta = 1 - alpha
img_2_with_reticle = np.copy(self.img2)
print "\n\n"
self._res_uv = dict()
# self._res_a_uv = dict()
# self._res_b_uv = dict()
for network_name in self._dcn_dict:
res_a = self._res_a[network_name]
res_b = self._res_b[network_name]
best_match_uv, best_match_diff, norm_diffs = \
DenseCorrespondenceNetwork.find_best_match((u, v), res_a, res_b)
print "\n\n"
print "network_name:", network_name
self._res_uv[network_name] = dict()
self._res_uv[network_name]['source'] = res_a[v, u, :].tolist()
self._res_uv[network_name]['target'] = res_b[v, u, :].tolist()
# print "res_a[v, u, :]:", res_a[v, u, :]
# print "res_b[v, u, :]:", res_b[v, u, :]
print "%s best match diff: %.3f" %(network_name, best_match_diff)
threshold = self._config["norm_diff_threshold"]
if network_name in self._config["norm_diff_threshold_dict"]:
threshold = self._config["norm_diff_threshold_dict"][network_name]
heatmap = self.scale_norm_diffs_to_make_heatmap(norm_diffs, threshold)
reticle_color = self._network_reticle_color[network_name]
draw_reticle(heatmap, best_match_uv[0], best_match_uv[1], reticle_color)
draw_reticle(img_2_with_reticle, best_match_uv[0], best_match_uv[1], reticle_color)
blended = cv2.addWeighted(self.img2_gray, alpha, heatmap, beta, 0)
cv2.imshow(network_name, blended)
cv2.imshow("target", img_2_with_reticle)
if event == cv2.EVENT_LBUTTONDOWN:
utils.saveToYaml(self._res_uv, 'clicked_point.yaml')
def save_configs(self):
"""
Saves config files to the logging directory
:return:
:rtype: None
"""
training_params_file = os.path.join(self._logging_dir, 'training.yaml')
utils.saveToYaml(self._config, training_params_file)
dataset_params_file = os.path.join(self._logging_dir, 'dataset.yaml')
utils.saveToYaml(self._dataset.config, dataset_params_file)
def save_configs(self):
"""
Saves config files to the logging directory
:return:
:rtype: None
"""
training_params_file = os.path.join(self._logging_dir, 'training.yaml')
utils.saveToYaml(self._config, training_params_file)
dataset_params_file = os.path.join(self._logging_dir, 'dataset.yaml')
utils.saveToYaml(self._dataset.config, dataset_params_file)
def save_network(self, dcn, optimizer, iteration, logging_dict=None):
"""
Saves network parameters to logging directory
:return:
:rtype: None
"""
network_param_file = os.path.join(self._logging_dir, utils.getPaddedString(iteration, width=6) + ".pth")
optimizer_param_file = network_param_file + ".opt"
torch.save(dcn.state_dict(), network_param_file)
torch.save(optimizer.state_dict(), optimizer_param_file)
# also save loss history stuff
if logging_dict is not None:
log_history_file = os.path.join(self._logging_dir, utils.getPaddedString(iteration, width=6) + "_log_history.yaml")
utils.saveToYaml(logging_dict, log_history_file)
current_loss_file = os.path.join(self._logging_dir, 'loss.yaml')
current_loss_data = self._get_current_loss(logging_dict)
utils.saveToYaml(current_loss_data, current_loss_file)
def save_configs(self):
"""
Saves config files to the logging directory
:return:
:rtype: None
"""
training_params_file = os.path.join(self._logging_dir, 'training.yaml')
utils.saveToYaml(self._config, training_params_file)
dataset_params_file = os.path.join(self._logging_dir, 'dataset.yaml')
utils.saveToYaml(self._dataset.config, dataset_params_file)
# make unique identifier
identifier_file = os.path.join(self._logging_dir, 'identifier.yaml')
identifier_dict = dict()
identifier_dict['id'] = utils.get_unique_string()
utils.saveToYaml(identifier_dict, identifier_file)
next_image_pair()
cv2.namedWindow('image1')
cv2.setMouseCallback('image1',draw_circle1)
cv2.namedWindow('image2')
cv2.setMouseCallback('image2',draw_circle2)
while(1):
cv2.imshow('image1',img1)
cv2.imshow('image2',img2)
k = cv2.waitKey(20) & 0xFF
if k == 27:
break
elif k == ord('a'):
print ix,iy
elif k == ord('s'):
if not check_same_length(img1_points_picked, img2_points_picked):
print "can't save when not same length"
print "try choosing a new image pair"
print "or picking more points on the one with less points"
else:
print "saving"
new_dict = make_savable_correspondence_pairs()
annotated_data.append(new_dict)
utils.saveToYaml(annotated_data, "new_annotated_pairs.yaml")
elif k == ord('n'):
next_image_pair()
cv2.destroyAllWindows()
next_image_pair()
cv2.namedWindow('image1')
cv2.setMouseCallback('image1', draw_circle1)
cv2.namedWindow('image2')
cv2.setMouseCallback('image2', draw_circle2)
while (1):
cv2.imshow('image1', img1)
cv2.imshow('image2', img2)
k = cv2.waitKey(20) & 0xFF
if k == 27:
break
elif k == ord('a'):
print(ix, iy)
elif k == ord('s'):
if not check_same_length(img1_points_picked, img2_points_picked):
print("can't save when not same length")
print("try choosing a new image pair")
print("or picking more points on the one with less points")
else:
print("saving")
new_dict = make_savable_correspondence_pairs()
annotated_data.append(new_dict)
utils.saveToYaml(annotated_data, "new_annotated_pairs.yaml")
elif k == ord('n'):
next_image_pair()
cv2.destroyAllWindows()
def find_best_match(self, event, u, v, flags, param):
"""
For each network, find the best match in the target image to point highlighted
with reticle in the source image. Displays the result
:return:
:rtype:
"""
if self._paused:
return
img_1_with_reticle = np.copy(self.img1)
draw_reticle(img_1_with_reticle, u, v, self._reticle_color)
cv2.imshow("source", img_1_with_reticle)
alpha = self._config["blend_weight_original_image"]
beta = 1 - alpha
img_2_with_reticle = np.copy(self.img2)
print("\n\n")
self._res_uv = dict()
# self._res_a_uv = dict()
# self._res_b_uv = dict()
for network_name in self._dcn_dict:
res_a = self._res_a[network_name]
res_b = self._res_b[network_name]
best_match_uv, best_match_diff, norm_diffs = \
DenseCorrespondenceNetwork.find_best_match((u, v), res_a, res_b)
print("\n\n")
print("network_name:", network_name)
print("scene_name_1", self._scene_name_1)
print("image_1_idx", self._image_1_idx)
print("scene_name_2", self._scene_name_2)
print("image_2_idx", self._image_2_idx)
d = dict()
d['scene_name'] = self._scene_name_1
d['image_idx'] = self._image_1_idx
d['descriptor'] = res_a[v, u, :].tolist()
d['u'] = u
d['v'] = v
print("\n-------keypoint info\n", d)
print("\n--------\n")
self._res_uv[network_name] = dict()
self._res_uv[network_name]['source'] = res_a[v, u, :].tolist()
self._res_uv[network_name]['target'] = res_b[v, u, :].tolist()
print("res_a[v, u, :]:", res_a[v, u, :])
print("res_b[v, u, :]:", res_b[best_match_uv[1],
best_match_uv[0], :])
print("%s best match diff: %.3f" % (network_name, best_match_diff))
print("res_a", self._res_uv[network_name]['source'])
print("res_b", self._res_uv[network_name]['target'])
threshold = self._config["norm_diff_threshold"]
if network_name in self._config["norm_diff_threshold_dict"]:
threshold = self._config["norm_diff_threshold_dict"][
network_name]
heatmap_color = vis_utils.compute_gaussian_kernel_heatmap_from_norm_diffs(
norm_diffs, self._config['kernel_variance'])
reticle_color = self._network_reticle_color[network_name]
draw_reticle(heatmap_color, best_match_uv[0], best_match_uv[1],
reticle_color)
draw_reticle(img_2_with_reticle, best_match_uv[0],
best_match_uv[1], reticle_color)
blended = cv2.addWeighted(self.img2, alpha, heatmap_color, beta, 0)
cv2.imshow(network_name, blended)
cv2.imshow("target", img_2_with_reticle)
if event == cv2.EVENT_LBUTTONDOWN:
utils.saveToYaml(self._res_uv, 'clicked_point.yaml')
def save_config(self, config_path=None):
if config_path is None:
utils.saveToYaml(self.config, self.config_path)
else:
utils.saveToYaml(self.config, config_path)
