def generate_labeled_image_list(self):
folders = glob.glob(self.db_base + '/*')
self.examples_all = []
for folder in folders:
id_ = self.id_class_mapping[os.path.basename(folder)[:7]]
folder_images = glob.glob(folder + IMAGE_FILE_ENDING)
if NUMBER_ALPHAS > 0:
alphas, _ = oal.read_fitting_log(folder + '/fitting.log')
for folder_image in folder_images:
example = Training_example()
example.images.append(folder_image)
example.xyz.append(folder_image.replace('isomap', 'xyzmap'))
example.label = id_
if NUMBER_ALPHAS > 0:
example.alphas.append(alphas)
self.examples_all.append(example)
if experiment[0] == num_imgs:
experiment.append(fitting_log)
#sort experiments by number of images
experiments.sort(key=lambda experiment: experiment[0])
for experiment in experiments:
print('for experiment with', experiment[0], 'images',
len(experiment) - 1, 'fittings have been found')
# now calculate the std deviation
experiment_std_devs = []
for experiment in experiments:
alphas = np.empty(((len(experiment) - 1), 63))
for idx, fitting in enumerate(experiment[1:]):
alphas[idx, :], _ = oal.read_fitting_log(fitting)
std_devs = np.std(alphas, axis=0)
experiment_std_devs.append(std_devs)
all_std_devs.append(experiment_std_devs)
#break
#print (len(all_std_devs), ",", len(all_std_devs[0]), ",", all_std_devs[0][0].shape )
#x_coordinates = [i[0] for i in experiments]
# make sure number of image sets is equal for all videos
min_num_sets = min([len(i) for i in all_std_devs
]) #if len(all_std_devs)>1 else len(all_std_devs[0])
#print([len(i) for i in all_std_devs])
#print (min_num_sets)
for video_idx in range(len(all_std_devs)):
all_std_devs[video_idx] = all_std_devs[video_idx][:min_num_sets]
# assemble all fitting results we find for this video and load the alphas
experiment_alphas = [[] for i in range(len(categories))]
#fitting_dirs = glob.glob(id_and_expression_dir+'/00[3,4,5,7]*')
fitting_dirs = glob.glob(id_and_expression_dir + '/pose_exp_*')
for fitting_dir in fitting_dirs:
pose_exp = int(fitting_dir.split('/')[-1][-5:-3])
iteration = int(fitting_dir.split('/')[-1][-2:])
fitting_log = fitting_dir + '/fitting.log'
if not os.path.exists(fitting_log):
print(
"ERROR: There is no fitting log file where there should be one!!",
fitting_dir)
exit(0)
alphas, angles = oal.read_fitting_log(fitting_log)
angles = np.array(angles)
# add the alphas to the specific experiment_alphas
experiment_alphas[pose_exp].append(alphas)
# print some information
for i in range(len(categories)):
print('for experiment with', categories[i], len(experiment_alphas[i]),
'fittings have been found that match the criteria')
# now calculate the std deviation
experiment_std_devs = []
for i in range(len(experiment_alphas)):
alphas = np.array(experiment_alphas[i])
std_devs = np.std(alphas, axis=0)
)
if len(id_and_expression_dirs) == 0:
print("ERROR: no videos found!!")
exit(0)
#categories = ['yaw<20', 'yaw>-20', '-30<yaw<30', '-40<yaw<40', 'yaw<-10 or yaw>10', 'all', '<-20 and +-20 and >20']
categories = ['exp ' + format(i, '01d') for i in range(10)]
all_errors = []
for id_and_expression_dir in id_and_expression_dirs:
print('id and expression dir: ', id_and_expression_dir)
# load "ground truth" alphas
gt_alphas, _ = oal.read_fitting_log(
GT_ALPHA_DIR + id_and_expression_dir.split('/')[-2] + "/" +
id_and_expression_dir.split('/')[-1][0:-5] + GT_ALPHA_EXP_FITTING_LOG)
# assemble all fitting results we find for this video and load the alphas
experiment_alphas = [[] for i in range(len(categories))]
#fitting_dirs = glob.glob(id_and_expression_dir+'/00[3,4,5,7]*')
fitting_dirs = glob.glob(id_and_expression_dir + '/pose_exp_*')
for fitting_dir in fitting_dirs:
pose_exp = int(fitting_dir.split('/')[-1][-5:-3])
iteration = int(fitting_dir.split('/')[-1][-2:])
fitting_log = fitting_dir + '/fitting.log'
if not os.path.exists(fitting_log):
print(
"ERROR: There is no fitting log file where there should be one!!",
fitting_dir)
def main(argv=None): # pylint: disable=unused-argument
if not os.path.exists(experiment_dir):
print('no experiment dir found!')
exit()
if not os.path.exists(train_dir):
print('no training dir found!')
exit()
if not os.path.exists(eval_dir):
print('no eval dir found!')
exit()
if not os.path.exists(eval_log):
print('no log file found!')
exit()
if not take_iter:
# find best network
best_accuracy = 0.0
best_iter = 0
with open(eval_log, 'r') as log:
for line in log:
iter_, accuracy = [float(x) for x in line.split()]
if accuracy > best_accuracy:
best_accuracy = accuracy
best_iter = int(iter_)
best_net_checkpoint = train_dir + '/model.ckpt-' + str(best_iter)
print('best network is', best_net_checkpoint)
# double check if we have this network checkpoint
if not os.path.exists(best_net_checkpoint + '.meta'):
print('shit! this checkpoint went missing... exiting ...')
exit(0)
else:
best_net_checkpoint = train_dir + '/model.ckpt-' + str(take_iter)
#db_loader = cnn_db_loader.PaSC_db_loader(db_base=PaSC_BASE, outputfolder=experiment_dir)
with tf.device('/gpu:0'):
if NUMBER_ALPHAS == 0:
vectors = test(best_net_checkpoint, image_list)
else:
print('reading fitting logs for alphas now')
all_alphas = []
old_alphas = []
old_folder = ''
for image in image_list:
folder = os.path.dirname(image)
if folder == old_folder:
alphas = old_alphas
else:
alphas, _ = oal.read_fitting_log(folder + '/fitting.log')
old_alphas = alphas
old_folder = folder
all_alphas.append(alphas)
vectors = test(best_net_checkpoint, image_list, alphas=all_alphas)
with open(test_log, 'w') as log:
for i in range(len(image_list)):
log.write(image_list[i] + ' ')
for x in range(vectors.shape[1]):
log.write(str(vectors[i, x]) + ' ')
log.write('\n')
OUTPUT_FILE = '/user/HS204/m09113/my_project_folder/PaSC/multi_fit_CCR_iter75_reg30_only_10_alphas_control_without_fte.csv'
QUERY_AND_TARGET_SAME = True
query = ET.parse(query_file)
query_root = query.getroot()
#print (query_root[0][0])
query_db = []
for i, query_video in enumerate(query_root):
query_video_name = query_video[0].attrib['file-name']
alphas = None
try:
query_alphas, _ = oal.read_fitting_log(FITTING_RESULTS_BASE +
query_video_name[:-4] +
'/fitting.log')
query_alphas = np.array(query_alphas)
#query_alphas = query_alphas/np.linalg.norm(query_alphas)
except oal.OalException:
print("No alphas found in", query_video_name)
except FileNotFoundError:
print("Video not found", query_video_name)
query_id_name = query_video.attrib['name']
query_db.append([query_id_name, query_alphas])
if i % 100 == 0:
print('loaded', i, 'of', len(query_root))
print('measuring distance...')
if QUERY_AND_TARGET_SAME:
target_db = query_db