def _get_data(self, descr, file_type, order=0):
if self.gt_scale == 1:
# original lowres version if gt_scale is 1
fname = "%s_%s.%s" % (descr, settings.LOWRES, file_type)
data = file_io.read_file(op.join(self.data_path, fname))
else:
# original highres version if gt_scale matches highres_scale
fname = "%s_%s.%s" % (descr, settings.HIGHRES, file_type)
data = file_io.read_file(op.join(self.data_path, fname))
# otherwise scale highres version to required shape
if self.gt_scale != self.highres_scale:
data = misc.resize_to_shape(data, self.get_height(), self.get_width(), order=order)
return data
def collect_scores(algorithms, scenes, metrics, masked=False):
scores_scenes_metrics_algos = np.full(
(len(scenes), len(metrics), len(algorithms)), fill_value=np.nan)
for idx_a, algorithm in enumerate(algorithms):
fname_json = op.join(settings.ALGO_EVAL_PATH, algorithm.get_name(),
"results.json")
try:
results = file_io.read_file(fname_json)
except IOError:
log.error(
"Could not find scores at: %s. \n"
"Please execute 'run_evaluation.py' with the algorithms, scenes and metrics "
"that you want to use in your figure." % fname_json)
exit()
for idx_s, scene in enumerate(scenes):
scene_scores = results[scene.get_name()]["scores"]
for idx_m, metric in enumerate(metrics):
metric_score = scene_scores.get(metric.get_id(), None)
if metric_score is not None:
scores_scenes_metrics_algos[idx_s, idx_m,
idx_a] = metric_score["value"]
if masked:
mask = get_mask_invalid(scores_scenes_metrics_algos)
scores_scenes_metrics_algos = np.ma.masked_array(
scores_scenes_metrics_algos, mask=mask)
return scores_scenes_metrics_algos
def initialize_algorithms(file_names_algorithms,
set_colors=True,
is_baseline=False,
is_meta=False):
try:
meta_data = file_io.read_file(settings.PATH_TO_ALGO_META_DATA)
except IOError:
meta_data = dict()
algorithms = []
for file_name in file_names_algorithms:
algorithm = Algorithm(file_name=file_name,
is_baseline=is_baseline,
is_meta=is_meta)
algo_data = meta_data.get(file_name, dict())
display_name = algo_data.get("acronym", None)
if display_name:
algorithm.display_name = display_name
algorithms.append(algorithm)
if set_colors:
algorithms = Algorithm.set_colors(algorithms)
return algorithms
def get_center_view(self):
fname = "input_Cam%03d.png" % self.get_center_cam()
center_view = file_io.read_file(op.join(self.data_path, fname))
if self.gt_scale != 1.0:
center_view = misc.resize_to_shape(center_view,
self.get_height(), self.get_width(), order=0)
return center_view
def main():
parser = OptionParser([
ConverterOps(input_help="path to png disparity map",
output_help="path to pfm disparity map")
])
image_path, config_path, pfm_path = parser.parse_args()
from toolkit.scenes import PhotorealisticScene
from toolkit.utils import log, file_io
import numpy as np
scene = PhotorealisticScene("demo", path_to_config=config_path)
disp_map = file_io.read_file(image_path)
log.info("Input range: [%0.1f, %0.1f]" %
(np.min(disp_map), np.max(disp_map)))
# scale from [MIN, MAX] to [disp_min, disp_max]
disp_map = (scene.disp_max - scene.disp_min) * (disp_map - MIN) / (
MAX - MIN) + scene.disp_min
log.info("Output range: [%0.1f, %0.1f]" %
(np.min(disp_map), np.max(disp_map)))
file_io.write_file(disp_map, pfm_path)
def main():
parser = OptionParser([ConverterOpsExt(input_help="path to disparity map",
output_help="path to point cloud",
optional_input=[("-c", "color_map_file",
"path to color map, "
"e.g. to center view of the scene")])])
disp_map_path, config_path, point_cloud_path, color_map_path = parser.parse_args()
from toolkit.scenes import PhotorealisticScene
from toolkit.utils import point_cloud, file_io
scene = PhotorealisticScene("demo", path_to_config=config_path)
disp_map = file_io.read_file(disp_map_path)
if color_map_path:
color_map = file_io.read_file(color_map_path)
else:
color_map = None
points = point_cloud.convert(scene, disp_map, color_map)
point_cloud.save(points, point_cloud_path)
def main():
parser = OptionParser([ConverterOps(input_help="path to disparity map",
output_help="path to depth map")])
disp_map_path, config_path, depth_map_path = parser.parse_args()
from toolkit.scenes import PhotorealisticScene
from toolkit.utils import file_io
scene = PhotorealisticScene("demo", path_to_config=config_path)
disp_map = file_io.read_file(disp_map_path)
depth_map = scene.disp2depth(disp_map)
file_io.write_file(depth_map, depth_map_path)
def evaluate(evaluation_output_path,
algorithm_input_path,
scenes,
metrics,
visualize=False,
add_to_existing_results=True,
add_pfms_to_result=True):
"""
:param evaluation_output_path: target directory for all evaluation results
:param algorithm_input_path: input directory for algorithm results,
expected directories: runtimes, disp_maps
:param scenes: scenes to be evaluated
:param metrics: metrics to be evaluated
:param visualize: whether to save visualizations (otherwise just the scores)
:param add_to_existing_results: if set to True, will try to read results.json and add/replace entries,
keeping existing scores of other scenes/metrics as is
:param add_pfms_to_result: when executed on evaluation server, pfms are prepared for 3D point cloud view
:return: success, {"messages": ["error 1", "error 2", ...]}
"""
log.info("Evaluating algorithm results in:\n %s" % algorithm_input_path)
log.info("Writing results to:\n %s" % evaluation_output_path)
log.info("Using ground truth data from:\n %s" % settings.DATA_PATH)
log.info("Metrics:\n %s" % ", ".join(m.get_display_name()
for m in metrics))
log.info("Scenes:\n %s" % ", ".join(s.get_display_name() for s in scenes))
file_name_results = op.join(evaluation_output_path, "results.json")
admin_errors = []
eval_json = dict()
if add_to_existing_results:
try:
eval_json = file_io.read_file(file_name_results)
except IOError:
pass
# evaluate
for scene in scenes:
scene_data = eval_json.get(scene.get_name(), dict())
try:
if visualize:
log.info("Visualizing algorithm result on %s" %
scene.get_display_name())
scene_data["algorithm_result"] = visualize_algo_result(
scene, algorithm_input_path, evaluation_output_path,
add_pfms_to_result)
log.info("Processing scene: %s" % scene.get_display_name())
log.info("Using data from:\n %s" % scene.get_data_path())
scene_scores = compute_scores(scene, metrics, algorithm_input_path,
evaluation_output_path, visualize)
if add_to_existing_results:
existing_scores = scene_data.get("scores", dict())
existing_scores.update(scene_scores)
scene_scores = existing_scores
scene_data["scores"] = scene_scores
except IOError as e:
admin_errors.append(e)
log.error(e)
continue
eval_json[scene.get_name()] = scene_data
# save json with scores and paths to visualizations
file_io.write_file(eval_json, file_name_results)
log.info("Done!")
success = not admin_errors
error_json = {"messages": admin_errors}
return success, error_json
def get_algo_result_from_dir(algo_dir, scene):
fname = get_fname_algo_result(algo_dir, scene)
algo_result = file_io.read_file(fname)
if scene.gt_scale != 1:
algo_result = sci.zoom(algo_result, scene.gt_scale, order=0)
return algo_result