def generate(args):
config = handle_arguments_and_configurations(name, args)
base_path = config['path']
if not os.path.isdir(base_path):
os.makedirs(base_path)
print("Generation in: {}".format(base_path))
images_path = os.path.join(base_path, "images")
if not os.path.isdir(images_path):
os.makedirs(images_path)
# Get configuration parameters.
vh_image_nbs = config['vh_image_nbs']
eye_diameter = config['eye_diameter']
mean_luminance = config['mean_luminance']
std_luminance = config['std_luminance']
normalized_value_median = config['normalized_value_median']
normalized_value_mad = config['normalized_value_mad']
display_rate = config['display_rate']
adaptation_duration = config['adaptation_duration']
flash_duration = config['flash_duration']
inter_flash_duration = config['inter_flash_duration']
frame_resolution = config['frame']['resolution']
frame_width = config['frame']['width']
frame_height = config['frame']['height']
nb_repetitions = config['nb_repetitions']
stuttering_vh_image_nbs = config['stuttering_vh_image_nbs']
nb_stuttering_vh_images = config['nb_stuttering_vh_images']
nb_stutters = config['nb_stutters']
seed = config['seed']
verbose = config['verbose']
# Fetch van Hateren images.
vh.fetch(image_nbs=vh_image_nbs,
download_if_missing=False,
verbose=verbose)
# Select unsaturated van Hateren image numbers.
if vh_image_nbs is None:
vh_image_nbs = vh.get_image_nbs()
else:
vh_image_nbs = np.array(vh_image_nbs)
are_saturated = vh.get_are_saturated(image_nbs=vh_image_nbs,
verbose=verbose)
are_unsaturated = np.logical_not(are_saturated)
assert vh_image_nbs.size == are_unsaturated.size, "{} != {}".format(
vh_image_nbs.size, are_unsaturated.size)
unsaturated_vh_image_nbs = vh_image_nbs[are_unsaturated]
# Select ...
# mean_luminances = vh.get_mean_luminances(image_nbs=unsaturated_vh_image_nbs, verbose=verbose)
# std_luminances = vh.get_std_luminances(image_nbs=unsaturated_vh_image_nbs, verbose=verbose)
max_luminances = vh.get_max_luminances(image_nbs=unsaturated_vh_image_nbs,
verbose=verbose)
# TODO remove the following lines?
# max_centered_luminances = max_luminances / mean_luminances
# # print(np.min(max_centered_luminances))
# # print(np.median(max_centered_luminances))
# # print(np.max(max_centered_luminances))
# are_good = max_centered_luminances <= 8.3581 # TODO correct?
# TODO remove the following lines?
# max_normalized_luminances = (max_luminances / mean_luminances - 1.0) / std_luminances + 1.0
# are_good = max_normalized_luminances <= 1.02
# TODO remove the following lines?
log_mean_luminances = vh.get_log_mean_luminances(
image_nbs=unsaturated_vh_image_nbs, verbose=verbose)
log_std_luminances = vh.get_log_mean_luminances(
image_nbs=unsaturated_vh_image_nbs, verbose=verbose)
log_max_luminances = np.log(1.0 + max_luminances)
log_max_normalized_luminances = (log_max_luminances -
log_mean_luminances) / log_std_luminances
are_good = log_max_normalized_luminances <= 5.0
# ...
good_vh_image_nbs = unsaturated_vh_image_nbs[are_good]
# ...
# selected_vh_image_nbs = unsaturated_vh_image_nbs
selected_vh_image_nbs = good_vh_image_nbs
# Check stuttering van Hateren image numbers.
np.random.seed(seed)
if stuttering_vh_image_nbs is None:
stuttering_vh_image_nbs = np.array([])
else:
assert len(stuttering_vh_image_nbs) <= nb_stuttering_vh_images
for stuttering_vh_image_nb in stuttering_vh_image_nbs:
assert stuttering_vh_image_nb in selected_vh_image_nbs, stuttering_vh_image_nb
potential_stuttering_vh_image_nbs = np.setdiff1d(selected_vh_image_nbs,
stuttering_vh_image_nbs,
assume_unique=True)
nb_missing_stuttering_vh_image_nbs = nb_stuttering_vh_images - len(
stuttering_vh_image_nbs)
stuttering_vh_image_nbs = np.concatenate(
(stuttering_vh_image_nbs,
np.random.choice(potential_stuttering_vh_image_nbs,
nb_missing_stuttering_vh_image_nbs,
replace=False)))
stuttering_vh_image_nbs.sort()
# Generate grey image.
image_filename = "image_{i:04d}.png".format(i=0)
image_path = os.path.join(images_path, image_filename)
# if not os.path.isfile(image_path): # TODO uncomment this line.
frame = get_grey_frame(frame_width, frame_height, luminance=mean_luminance)
frame = float_frame_to_uint8_frame(frame)
image = create_png_image(frame)
image.save(image_path)
# Extract images from the van Hateren images.
for vh_image_nb in tqdm.tqdm(selected_vh_image_nbs):
# Check if image already exists.
image_filename = "image_{i:04d}.png".format(i=vh_image_nb)
image_path = os.path.join(images_path, image_filename)
if os.path.isfile(image_path):
continue
# Cut out central sub-region.
a_x = vh.get_horizontal_angles()
a_y = vh.get_vertical_angles()
luminance_data = vh.load_luminance_data(vh_image_nb)
rbs = sp.interpolate.RectBivariateSpline(a_x,
a_y,
luminance_data,
kx=1,
ky=1)
angular_resolution = math.atan(
frame_resolution / eye_diameter) * (180.0 / math.pi)
a_x = compute_horizontal_angles(width=frame_width,
angular_resolution=angular_resolution)
a_y = compute_vertical_angles(height=frame_height,
angular_resolution=angular_resolution)
a_x, a_y = np.meshgrid(a_x, a_y)
luminance_data = rbs.ev(a_x, a_y)
luminance_data = luminance_data.transpose()
# TODO uncomment the following lines?
# # Prepare data.
# luminance_median = np.median(luminance_data)
# luminance_mad = np.median(np.abs(luminance_data - luminance_median))
# centered_luminance_data = luminance_data - luminance_median
# if luminance_mad > 0.0:
# centered_n_reduced_luminance_data = centered_luminance_data / luminance_mad
# else:
# centered_n_reduced_luminance_data = centered_luminance_data
# normalized_luminance_data = centered_n_reduced_luminance_data
# scaled_data = normalized_value_mad * normalized_luminance_data
# shifted_n_scaled_data = scaled_data + normalized_value_median
# data = shifted_n_scaled_data
# TODO remove the 2 following lines?
# scaled_luminance_data = luminance_data / np.mean(luminance_data)
# data = scaled_luminance_data / 8.3581
# TODO remove the 2 following lines?
# normalized_luminance_data = (luminance_data / np.mean(luminance_data) - 1.0) / np.std(luminance_data) + 1.0
# data = (normalized_luminance_data - 1.0) / 0.02 * 0.8 + 0.2
# TODO remove the following lines?
log_luminance_data = np.log(1.0 + luminance_data)
log_mean_luminance = np.mean(log_luminance_data)
log_std_luminance = np.std(log_luminance_data)
normalized_log_luminance_data = log_luminance_data - log_mean_luminance
if log_std_luminance > 1e-13:
normalized_log_luminance_data = normalized_log_luminance_data / log_std_luminance
normalized_log_luminance_data = 0.2 * normalized_log_luminance_data
normalized_luminance_data = np.exp(normalized_log_luminance_data) - 1.0
normalized_luminance_data = normalized_luminance_data - np.mean(
normalized_luminance_data)
if np.std(normalized_luminance_data) > 1e-13:
normalized_luminance_data = normalized_luminance_data / np.std(
normalized_luminance_data)
data = std_luminance * normalized_luminance_data + mean_luminance
# Prepare image data
if np.count_nonzero(data < 0.0) > 0:
s = "some pixels are negative in image {} (consider changing the configuration, 'normalized_value_mad': {})"
message = s.format(
vh_image_nb, normalized_value_mad /
((normalized_value_median - np.min(data)) /
(normalized_value_median - 0.0)))
warnings.warn(message)
data[data < 0.0] = 0.0
if np.count_nonzero(data > 1.0) > 0:
s = "some pixels saturate in image {} (consider changing the configuration, 'normalized_value_mad': {})"
message = s.format(
vh_image_nb, normalized_value_mad /
((np.max(data) - normalized_value_median) /
(1.0 - normalized_value_median)))
warnings.warn(message)
data[data > 1.0] = 1.0
data = np.array(
254.0 * data,
dtype=np.uint8) # 0.0 -> 0 and 1.0 -> 254 such that 0.5 -> 127
data = np.transpose(data)
data = np.flipud(data)
image = create_png_image(data)
# Save image.
image.save(image_path)
# Set condition numbers and image paths.
condition_nbs = []
stuttering_condition_nbs = []
image_paths = {}
for k, vh_image_nb in enumerate(selected_vh_image_nbs):
# condition_nb = k + 1
condition_nb = k
image_filename = 'image_{i:04d}.png'.format(i=vh_image_nb)
image_path = os.path.join(images_path, image_filename)
assert condition_nb not in condition_nbs
assert os.path.isfile(image_path)
condition_nbs.append(condition_nb)
if vh_image_nb in stuttering_vh_image_nbs:
stuttering_condition_nbs.append(condition_nb)
image_paths[condition_nb] = image_path
condition_nbs = np.array(condition_nbs)
stuttering_condition_nbs = np.array(stuttering_condition_nbs)
nb_conditions = len(condition_nbs)
nb_stuttering_conditions = len(stuttering_condition_nbs)
# Create conditions .csv file.
conditions_csv_filename = '{}_conditions.csv'.format(name)
conditions_csv_path = os.path.join(base_path, conditions_csv_filename)
print("Start creating conditions .csv file...")
# Open conditions .csv file.
conditions_csv_file = open_csv_file(conditions_csv_path, columns=['path'])
# Add conditions for van Hateren images.
for condition_nb in condition_nbs:
image_path = image_paths[condition_nb]
path = image_path.replace(base_path, '')
path = path[1:] # remove separator
conditions_csv_file.append(path=path)
# Close conditions .csv file.
conditions_csv_file.close()
# ...
print("End of conditions .csv file creation.")
# Set sequence of conditions for each repetition.
repetition_sequences = {}
np.random.seed(seed)
normal_condition_nbs = np.setdiff1d(condition_nbs,
stuttering_condition_nbs,
assume_unique=True)
nb_normal_indices = len(normal_condition_nbs)
nb_stuttering_indices = nb_stuttering_conditions * nb_stutters
nb_indices = nb_normal_indices + nb_stuttering_indices
sequence = np.empty(nb_indices, dtype=np.int)
stuttering_indices = np.linspace(0,
nb_indices,
num=nb_stuttering_indices,
endpoint=False)
stuttering_indices = stuttering_indices.astype(np.int)
normal_indices = np.setdiff1d(np.arange(0, nb_indices), stuttering_indices)
sequence[stuttering_indices] = np.concatenate(
tuple([stuttering_condition_nbs for _ in range(0, nb_stutters)]))
for repetition_nb in range(0, nb_repetitions):
repetition_sequence = np.copy(sequence)
# Normal.
repetition_normal_sequence = np.copy(normal_condition_nbs)
np.random.shuffle(repetition_normal_sequence)
repetition_sequence[normal_indices] = repetition_normal_sequence
# Stuttering.
repetition_stuttering_sequence = []
for _ in range(0, nb_stutters):
repetition_stuttering_condition_nbs = np.copy(
stuttering_condition_nbs)
np.random.shuffle(repetition_stuttering_condition_nbs)
repetition_stuttering_sequence.append(
repetition_stuttering_condition_nbs)
repetition_stuttering_sequence = np.concatenate(
tuple(repetition_stuttering_sequence))
repetition_sequence[
stuttering_indices] = repetition_stuttering_sequence
# ...
repetition_sequences[repetition_nb] = repetition_sequence
# Create .bin file.
print("Start creating .bin file...")
bin_filename = '{}.bin'.format(name)
bin_path = os.path.join(base_path, bin_filename)
nb_bin_images = 1 + nb_conditions # i.e. grey image and other conditions
bin_frame_nbs = {}
# Open .bin file.
bin_file = open_bin_file(bin_path,
nb_bin_images,
frame_width=frame_width,
frame_height=frame_height,
reverse=False,
mode='w')
# Add grey frame.
grey_frame = get_grey_frame(frame_width,
frame_height,
luminance=mean_luminance)
grey_frame = float_frame_to_uint8_frame(grey_frame)
bin_file.append(grey_frame)
bin_frame_nbs[None] = bin_file.get_frame_nb()
# Add van Hateren frames.
for condition_nb in tqdm.tqdm(condition_nbs):
image_path = image_paths[condition_nb]
image = load_png_image(image_path)
frame = image.data
bin_file.append(frame)
bin_frame_nbs[condition_nb] = bin_file.get_frame_nb()
# Close .bin file.
bin_file.close()
# ...
print("End of .bin file creation.")
# Create .vec file and stimulation .csv file.
print("Start creating .vec file...")
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(base_path, vec_filename)
csv_filename = "{}_trials.csv".format(name)
csv_path = os.path.join(base_path, csv_filename)
# ...
nb_displays_during_adaptation = int(
np.ceil(adaptation_duration * display_rate))
nb_displays_per_flash = int(np.ceil(flash_duration * display_rate))
nb_displays_per_inter_flash = int(
np.ceil(inter_flash_duration * display_rate))
nb_flashes_per_repetition = nb_conditions + nb_stuttering_vh_images * (
nb_stutters - 1)
nb_displays_per_repetition = nb_flashes_per_repetition * (
nb_displays_per_flash + nb_displays_per_inter_flash)
nb_displays = nb_displays_during_adaptation + nb_repetitions * nb_displays_per_repetition
# Open .vec file.
vec_file = open_vec_file(vec_path, nb_displays=nb_displays)
# Open .csv file.
csv_file = open_csv_file(
csv_path,
columns=['condition_nb', 'start_display_nb', 'end_display_nb'])
# Add adaptation.
bin_frame_nb = bin_frame_nbs[None] # i.e. default frame (grey)
for _ in range(0, nb_displays_during_adaptation):
vec_file.append(bin_frame_nb)
# Add repetitions.
for repetition_nb in tqdm.tqdm(range(0, nb_repetitions)):
repetition_sequence = repetition_sequences[repetition_nb]
for condition_nb in repetition_sequence:
# Add flash.
start_display_nb = vec_file.get_display_nb() + 1
bin_frame_nb = bin_frame_nbs[condition_nb]
for _ in range(0, nb_displays_per_flash):
vec_file.append(bin_frame_nb)
end_display_nb = vec_file.get_display_nb()
csv_file.append(condition_nb=condition_nb,
start_display_nb=start_display_nb,
end_display_nb=end_display_nb)
# Add inter flash.
bin_frame_nb = bin_frame_nbs[None] # i.e. default frame (grey)
for _ in range(0, nb_displays_per_inter_flash):
vec_file.append(bin_frame_nb)
# Close .csv file.
csv_file.close()
# Close .vec file.
vec_file.close()
# ...
print("End of .vec file creation.")
return
def generate(args):
config = handle_arguments_and_configurations(name, args)
base_path = config['path']
if not os.path.isdir(base_path):
os.makedirs(base_path)
print("Generation in {}".format(base_path))
# Create directories (if necessary).
images_dirname = 'images'
images_path = os.path.join(base_path, images_dirname)
if not os.path.isdir(images_path):
os.makedirs(images_path)
patterns_dirname = 'patterns'
patterns_path = os.path.join(base_path, patterns_dirname)
if not os.path.isdir(patterns_path):
os.makedirs(patterns_path)
frames_dirname = 'frames'
frames_path = os.path.join(base_path, frames_dirname)
if not os.path.isdir(frames_path):
os.makedirs(frames_path)
# Get configuration parameters.
image_keys = config['images']
pattern_nbs = config['perturbations']['pattern_nbs']
amplitude_value = config['perturbations']['amplitude']
eye_diameter = config['eye_diameter']
mean_luminance = config['mean_luminance']
std_luminance = config['std_luminance']
display_rate = config['display_rate']
adaptation_duration = config['adaptation_duration']
flash_duration = config['flash_duration']
inter_flash_duration = config['inter_flash_duration']
frame_width = config['frame']['width']
frame_height = config['frame']['height']
frame_resolution = config['frame']['resolution']
nb_repetitions = config['nb_repetitions']
seed = config['seed']
# Fetch images.
image_nbs = np.array(list(image_keys.keys()), dtype=int)
for image_nb in image_nbs:
image_key = image_keys[str(image_nb)]
fetch_image(*image_key)
nb_images = len(image_nbs)
_ = nb_images
# Fetch patterns.
pattern_nbs = np.array(pattern_nbs)
fetch_patterns(image_nbs=pattern_nbs)
nb_patterns = len(pattern_nbs)
# Prepare image parameters.
images_params = collections.OrderedDict()
for image_nb in image_nbs:
filename = 'image_{nb:01d}_data.npy'.format(nb=image_nb)
path = os.path.join(images_dirname, filename)
images_params[image_nb] = collections.OrderedDict([('path', path)])
# Prepare pattern parameters.
patterns_params = collections.OrderedDict()
for pattern_nb in pattern_nbs:
filename = 'pattern_{nb:01d}_data.npy'.format(nb=pattern_nb)
path = os.path.join(patterns_dirname, filename)
patterns_params[pattern_nb] = collections.OrderedDict([('path', path)])
def get_image_data(image_nb):
# Load image data.
image_key = image_keys[str(image_nb)]
dataset_name = image_key[0]
dataset = get_dataset(dataset_name)
data = dataset.load_data(*image_key[1:])
# Cut out central sub-regions.
a_x = dataset.get_horizontal_angles()
a_y = dataset.get_vertical_angles()
rbs = sp.interpolate.RectBivariateSpline(a_x, a_y, data)
angular_resolution = math.atan(
frame_resolution / eye_diameter) * (180.0 / math.pi)
a_x = compute_horizontal_angles(width=frame_width,
angular_resolution=angular_resolution)
a_y = compute_vertical_angles(height=frame_height,
angular_resolution=angular_resolution)
a_x, a_y = np.meshgrid(a_x, a_y)
data = rbs.ev(a_x, a_y)
data = data.transpose()
# # Prepare image data.
# median = np.median(data)
# mad = np.median(np.abs(data - median))
# # mad = np.std(data)
# centered_data = data - median
# if mad > 1.e-13:
# centered_n_reduced_data = centered_data / mad
# else:
# centered_n_reduced_data = centered_data
# normalized_data = centered_n_reduced_data
# scaled_data = 0.1 * normalized_data
# shifted_n_scaled_data = scaled_data + 0.5
# TODO keep the following normalization?
# # Prepare image data.
# mean = np.mean(data)
# scaled_data = data / mean if mean > 0.0 else data
# shifted_n_scaled_data = 0.2 * scaled_data # TODO correct?
# TODO keep the following normalization?
luminance_data = data
log_luminance_data = np.log(1.0 + luminance_data)
log_mean_luminance = np.mean(log_luminance_data)
log_std_luminance = np.std(log_luminance_data)
normalized_log_luminance_data = log_luminance_data - log_mean_luminance
if log_std_luminance > 1e-13:
normalized_log_luminance_data = normalized_log_luminance_data / log_std_luminance
normalized_log_luminance_data = 0.2 * normalized_log_luminance_data
normalized_luminance_data = np.exp(normalized_log_luminance_data) - 1.0
normalized_luminance_data = normalized_luminance_data - np.mean(
normalized_luminance_data)
if np.std(normalized_luminance_data) > 1e-13:
normalized_luminance_data = normalized_luminance_data / np.std(
normalized_luminance_data)
luminance_data = std_luminance * normalized_luminance_data + mean_luminance
# Save image data.
image_data_path = os.path.join(base_path,
images_params[image_nb]['path'])
np.save(image_data_path, luminance_data)
# Prepare image.
data = np.copy(luminance_data)
data[data < 0.0] = 0.0
data[data > 1.0] = 1.0
data = np.array(
254.0 * data,
dtype=np.uint8) # 0.0 -> 0 and 1.0 -> 254 such that 0.5 -> 127
data = np.transpose(data)
data = np.flipud(data)
image = create_png_image(data)
# Save image.
image_image_filename = "image_{nb:01d}_image.png".format(nb=image_nb)
image_image_path = os.path.join(images_path, image_image_filename)
image.save(image_image_path)
return luminance_data
def get_pattern_data(pattern_nb):
# Load pattern data.
data = load_checkerboard_data(pattern_nb, with_borders=0.5)
data = 2.0 * (data - 0.5) / 254.0
# Save pattern data.
pattern_data_path = os.path.join(base_path,
patterns_params[pattern_nb]['path'])
np.save(pattern_data_path, data[1:-1, 1:-1]) # without borders
# Project.
# a_x = cb.get_horizontal_angles()
# a_y = cb.get_vertical_angles()
# rbs = sp.interpolate.RectBivariateSpline(a_x, a_y, data, kx=1, ky=1)
a_x = cb.get_horizontal_angles(with_borders=True)
a_y = cb.get_vertical_angles(with_borders=True)
a_x, a_y = np.meshgrid(a_x, a_y)
a = np.stack((np.ravel(a_x), np.ravel(a_y)),
axis=1) # i.e. stack along 2nd axis
ni = sp.interpolate.NearestNDInterpolator(a, np.ravel(data))
angular_resolution = math.atan(
frame_resolution / eye_diameter) * (180.0 / math.pi)
a_x = compute_horizontal_angles(width=frame_width,
angular_resolution=angular_resolution)
a_y = compute_vertical_angles(height=frame_height,
angular_resolution=angular_resolution)
a_x, a_y = np.meshgrid(a_x, a_y)
# data = rbs.ev(a_x, a_y)
# data = data.transpose()
a = np.stack((np.ravel(a_x), np.ravel(a_y)),
axis=1) # i.e. stack along 2nd axis
data = ni(a)
shape = (frame_width, frame_height)
data = np.reshape(data, shape)
data = data.transpose()
# Create pattern image.
image_data = data
image_data = 254.0 * image_data
image_data = np.array(
image_data,
dtype=np.uint8) # 0.0 -> 0 and 1.0 -> 254 such that 0.5 -> 127
image_data = np.transpose(image_data)
image_data = np.flipud(image_data)
pattern = create_png_image(image_data)
# Save pattern.
pattern_image_filename = "pattern_{nb:01d}_image.png".format(
nb=pattern_nb)
pattern_image_path = os.path.join(patterns_path,
pattern_image_filename)
pattern.save(pattern_image_path)
return data
def get_frame_path(image_nb, pattern_nb):
image_index = np.where(image_nbs == image_nb)[0][0]
pattern_index = np.where(pattern_nbs == pattern_nb)[0][0]
frame_nb = (image_index * nb_patterns) + pattern_index
filename = "frame_{nb:04d}.png".format(nb=frame_nb)
path = os.path.join(frames_path, filename)
return path
# Set condition parameters.
condition_nb = 0
conditions_params = collections.OrderedDict()
frame_paths = collections.OrderedDict()
for image_nb in image_nbs:
for pattern_nb in pattern_nbs:
assert condition_nb not in conditions_params
conditions_params[condition_nb] = collections.OrderedDict([
('image_nb', image_nb),
('pattern_nb', pattern_nb),
])
frame_paths[condition_nb] = get_frame_path(image_nb, pattern_nb)
condition_nb += 1
condition_nbs = np.array(list(conditions_params.keys()))
nb_conditions = len(condition_nbs)
# Create frames.
# # Preload images data.
images_data = {}
for image_nb in image_nbs:
# Get image data.
image_data = get_image_data(image_nb)
# Store image data.
images_data[image_nb] = image_data
# # Create frames.
pattern_data = None
for pattern_nb in tqdm.tqdm(pattern_nbs):
for image_nb in image_nbs:
frame_path = get_frame_path(image_nb, pattern_nb)
if os.path.isfile(frame_path):
continue
# Get image data.
image_data = images_data[image_nb]
# Get pattern data.
pattern_data = get_pattern_data(
pattern_nb) if pattern_data is None else pattern_data
# Create frame data.
data = image_data + amplitude_value * pattern_data
# Create frame image.
data[data < 0.0] = 0.0
data[data > 1.0] = 1.0
data = np.array(
254.0 * data,
dtype=np.uint8) # 0.0 -> 0 and 1.0 -> 254 such that 0.5 -> 127
data = np.transpose(data)
data = np.flipud(data)
image = create_png_image(data)
# Save frame image.
image.save(frame_path)
pattern_data = None
# Set image ordering for each repetitions.
repetition_orderings = collections.OrderedDict()
np.random.seed(seed)
for repetition_nb in range(0, nb_repetitions):
ordering = np.copy(condition_nbs)
np.random.shuffle(ordering)
repetition_orderings[repetition_nb] = ordering
# Create conditions .csv file.
conditions_csv_filename = '{}_conditions.csv'.format(name)
conditions_csv_path = os.path.join(base_path, conditions_csv_filename)
print("Start creating conditions .csv file...")
conditions_csv_file = open_csv_file(conditions_csv_path,
columns=['image_nb', 'pattern_nb'])
for condition_nb in condition_nbs:
condition_params = conditions_params[condition_nb]
conditions_csv_file.append(**condition_params)
conditions_csv_file.close()
print("End of conditions .csv file creation.")
# Create images .csv file.
images_csv_filename = '{}_images.csv'.format(name)
images_csv_path = os.path.join(base_path, images_csv_filename)
print("Start creating images .csv file...")
images_csv_file = open_csv_file(images_csv_path, columns=['path'])
for image_nb in image_nbs:
image_params = images_params[image_nb]
images_csv_file.append(**image_params)
images_csv_file.close()
print("End of images .csv file creation.")
# Create patterns .csv file.
patterns_csv_filename = '{}_patterns.csv'.format(name)
patterns_csv_path = os.path.join(base_path, patterns_csv_filename)
print("Start creating patterns .csv file...")
patterns_csv_file = open_csv_file(patterns_csv_path, columns=['path'])
for pattern_nb in pattern_nbs:
pattern_params = patterns_params[pattern_nb]
patterns_csv_file.append(**pattern_params)
patterns_csv_file.close()
print("End of patterns .csv file creation.")
# Create .bin file.
print("Start creating .bin file...")
bin_filename = '{}.bin'.format(name)
bin_path = os.path.join(base_path, bin_filename)
nb_bin_images = 1 + nb_conditions # i.e. grey image and other conditions
bin_frame_nbs = {}
# Open .bin file.
bin_file = open_bin_file(bin_path,
nb_bin_images,
frame_width=frame_width,
frame_height=frame_height,
reverse=False,
mode='w')
# Add grey frame.
grey_frame = get_grey_frame(frame_width,
frame_height,
luminance=mean_luminance)
grey_frame = float_frame_to_uint8_frame(grey_frame)
bin_file.append(grey_frame)
bin_frame_nbs[None] = bin_file.get_frame_nb()
# Add frames.
for condition_nb in tqdm.tqdm(condition_nbs):
frame_path = frame_paths[condition_nb]
frame = load_png_image(frame_path)
bin_file.append(frame.data)
bin_frame_nbs[condition_nb] = bin_file.get_frame_nb()
# Close .bin file.
bin_file.close()
# ...
print("End of .bin file creation.")
# Create .vec file.
print("Start creating .vec file...")
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(base_path, vec_filename)
csv_filename = "{}_trials.csv".format(name)
csv_path = os.path.join(base_path, csv_filename)
# ...
nb_displays_during_adaptation = int(
np.ceil(adaptation_duration * display_rate))
nb_displays_per_flash = int(np.ceil(flash_duration * display_rate))
nb_displays_per_inter_flash = int(
np.ceil(inter_flash_duration * display_rate))
nb_displays_per_repetition = nb_conditions * (nb_displays_per_flash +
nb_displays_per_inter_flash)
nb_displays = nb_displays_during_adaptation + nb_repetitions * nb_displays_per_repetition
# Open .vec file.
vec_file = open_vec_file(vec_path, nb_displays=nb_displays)
# Open .csv file.
csv_file = open_csv_file(
csv_path,
columns=['condition_nb', 'start_display_nb', 'end_display_nb'])
# Add adaptation.
bin_frame_nb = bin_frame_nbs[None] # i.e. default frame (grey)
for _ in range(0, nb_displays_during_adaptation):
vec_file.append(bin_frame_nb)
# Add repetitions.
for repetition_nb in tqdm.tqdm(range(0, nb_repetitions)):
condition_nbs = repetition_orderings[repetition_nb]
for condition_nb in condition_nbs:
# Add flash.
start_display_nb = vec_file.get_display_nb() + 1
bin_frame_nb = bin_frame_nbs[condition_nb]
for _ in range(0, nb_displays_per_flash):
vec_file.append(bin_frame_nb)
end_display_nb = vec_file.get_display_nb()
csv_file.append(condition_nb=condition_nb,
start_display_nb=start_display_nb,
end_display_nb=end_display_nb)
# Add inter flash.
bin_frame_nb = bin_frame_nbs[None] # i.e. default frame (grey)
for _ in range(0, nb_displays_per_inter_flash):
vec_file.append(bin_frame_nb)
# Close .csv file.
csv_file.close()
# Close .vec file.
vec_file.close()
# ...
print("End of .vec file creation.")
# TODO add unperturbed flashed images?
return
def generate(args):
config = handle_arguments_and_configurations(name, args)
path = config['path']
if not os.path.isdir(path):
os.makedirs(path)
print(path)
reference_images_path = os.path.join(path, "reference_images")
if not os.path.isdir(reference_images_path):
os.makedirs(reference_images_path)
perturbation_patterns_path = os.path.join(path, "perturbation_patterns")
if not os.path.isdir(perturbation_patterns_path):
os.makedirs(perturbation_patterns_path)
frames_path = os.path.join(path, "frames")
if not os.path.isdir(frames_path):
os.makedirs(frames_path)
# Get configuration parameters.
reference_images = config['reference_images']
nb_horizontal_checks = config['perturbations']['nb_horizontal_checks']
nb_vertical_checks = config['perturbations']['nb_vertical_checks']
assert nb_horizontal_checks % 2 == 0, "number of checks should be even (horizontally): {}".format(
nb_horizontal_checks)
assert nb_vertical_checks % 2 == 0, "number of checks should be even (vertically): {}".format(
nb_vertical_checks)
with_random_patterns = config['perturbations']['with_random_patterns']
perturbation_patterns_indices = config['perturbations']['pattern_indices']
perturbation_amplitudes = config['perturbations']['amplitudes']
perturbation_amplitudes_indices = [
k for k, _ in enumerate(perturbation_amplitudes)
]
display_rate = config['display_rate']
frame_width_in_px = config['frame']['width']
frame_height_in_px = config['frame']['height']
frame_duration = config['frame']['duration']
nb_repetitions = config['nb_repetitions']
# Collect reference images.
reference_indices = [int(key) for key in reference_images.keys()]
for reference_index in reference_indices:
dataset, index = reference_images[str(reference_index)]
collect_reference_image(reference_index,
dataset=dataset,
index=index,
path=reference_images_path,
config=config)
# Create .csv file for reference_image.
csv_filename = "{}_reference_images.csv".format(name)
csv_path = os.path.join(path, csv_filename)
columns = ['reference_image_path']
csv_file = open_csv_file(csv_path, columns=columns)
for index in reference_indices:
reference_image_path = os.path.join(
"reference_images", "reference_{:05d}.png".format(index))
csv_file.append(reference_image_path=reference_image_path)
csv_file.close()
# Prepare perturbation pattern indices.
if with_random_patterns:
# TODO check the following lines!
# Compute the number of random patterns.
nb_perturbation_patterns = len(perturbation_patterns_indices)
nb_perturbation_amplitudes = len(perturbation_amplitudes_indices)
nb_perturbations = nb_perturbation_patterns * nb_perturbation_amplitudes
nb_random_patterns = nb_perturbations * nb_repetitions
print("number of random patterns: {}".format(
nb_random_patterns)) # TODO remove this line.
# Choose the indices of the random patterns.
random_patterns_indices = nb_perturbation_patterns + np.arange(
0, nb_random_patterns)
# Define pattern indices.
all_patterns_indices = np.concatenate(
(perturbation_patterns_indices, random_patterns_indices))
else:
nb_random_patterns = 0
random_patterns_indices = None
all_patterns_indices = perturbation_patterns_indices
print("Start collecting perturbation patterns...")
# TODO remove the following commented lines?
# for index in perturbation_patterns_indices:
# collect_perturbation_pattern(index, nb_horizontal_checks=nb_horizontal_checks,
# nb_vertical_checks=nb_vertical_checks, path=perturbation_patterns_path)
for index in tqdm.tqdm(all_patterns_indices):
collect_perturbation_pattern(index,
nb_horizontal_checks=nb_horizontal_checks,
nb_vertical_checks=nb_vertical_checks,
path=perturbation_patterns_path)
print("End collecting perturbation patterns.")
# Create .csv file for perturbation pattern.
csv_filename = "{}_perturbation_patterns.csv".format(name)
csv_path = os.path.join(path, csv_filename)
columns = ['perturbation_pattern_path']
csv_file = open_csv_file(csv_path, columns=columns)
# TODO remove the following commented line?
# for index in perturbation_patterns_indices:
for index in all_patterns_indices:
perturbation_pattern_path = os.path.join(
"perturbation_patterns", "checkerboard{:05d}.png".format(index))
csv_file.append(perturbation_pattern_path=perturbation_pattern_path)
csv_file.close()
# Create .csv file for perturbation amplitudes.
csv_filename = "{}_perturbation_amplitudes.csv".format(name)
csv_path = os.path.join(path, csv_filename)
columns = ['perturbation_amplitude']
csv_file = open_csv_file(csv_path, columns=columns)
for perturbation_amplitude in perturbation_amplitudes:
csv_file.append(perturbation_amplitude=perturbation_amplitude)
csv_file.close()
# Compute the number of images.
nb_reference_images = len(reference_indices)
nb_perturbation_patterns = len(perturbation_patterns_indices)
nb_perturbation_amplitudes = len(perturbation_amplitudes_indices)
nb_images = 1 + nb_reference_images * (
1 + nb_perturbation_patterns * nb_perturbation_amplitudes)
if with_random_patterns:
nb_images = nb_images + nb_reference_images * nb_random_patterns # TODO check this line!
combinations = get_combinations(reference_indices,
perturbation_patterns_indices,
perturbation_amplitudes_indices)
if with_random_patterns:
nb_deterministic_combinations = len(
combinations) + 1 # +1 to take inter-stimulus frame into account
random_combinations = get_random_combinations(
reference_indices, random_patterns_indices,
nb_deterministic_combinations)
else:
random_combinations = None
# Create .csv file.
csv_filename = "{}_combinations.csv".format(name)
csv_path = os.path.join(path, csv_filename)
columns = [
'reference_id', 'perturbation_pattern_id', 'perturbation_amplitude_id'
]
csv_file = open_csv_file(csv_path, columns=columns, dtype='Int64')
for combination_index in combinations:
combination = combinations[combination_index]
# TODO remove the following commented lines?
# kwargs = {
# 'reference_id': reference_indices[combination[0]],
# 'perturbation_pattern_id': perturbation_patterns_indices[combination[1]],
# 'perturbation_amplitude_id': perturbation_amplitudes_indices[combination[2]],
# }
reference_id, perturbation_pattern_id, perturbation_amplitude_id = combination
kwargs = {
'reference_id': reference_id,
'perturbation_pattern_id': perturbation_pattern_id,
'perturbation_amplitude_id': perturbation_amplitude_id,
}
csv_file.append(**kwargs)
# TODO check the following lines!
# Add random pattern (if necessary).
if with_random_patterns:
for combination_index in random_combinations:
combination = random_combinations[combination_index]
reference_id, pattern_id, amplitude_id = combination
kwargs = {
'reference_id': reference_id,
'perturbation_pattern_id': pattern_id,
'perturbation_amplitude_id': amplitude_id,
}
csv_file.append(**kwargs)
csv_file.close()
# TODO fix the permutations?
nb_combinations = len(combinations)
nb_frame_displays = int(display_rate * frame_duration)
assert display_rate * frame_duration == float(nb_frame_displays)
nb_displays = nb_frame_displays + nb_repetitions * nb_combinations * (
2 * nb_frame_displays)
if with_random_patterns:
nb_random_combinations = len(random_combinations)
nb_displays = nb_displays + nb_random_combinations * (
2 * nb_frame_displays)
display_time = float(nb_displays) / display_rate
print("display time: {} s ({} min)".format(display_time,
display_time / 60.0))
# TODO improve feedback.
combination_indices = list(combinations)
permutations = get_permutations(combination_indices,
nb_repetitions=nb_repetitions)
if with_random_patterns:
random_combination_indices = list(random_combinations)
random_combination_groups = get_random_combination_groups(
random_combination_indices, nb_repetitions=nb_repetitions)
else:
random_combination_groups = None
print("Start creating .bin file...")
# Create .bin file.
bin_filename = "fipwc.bin"
bin_path = os.path.join(path, bin_filename)
bin_file = open_bin_file(bin_path,
nb_images,
frame_width=frame_width_in_px,
frame_height=frame_height_in_px)
# Save grey frame.
grey_frame = get_grey_frame(frame_width_in_px,
frame_height_in_px,
luminance=0.5)
grey_frame = float_frame_to_uint8_frame(grey_frame)
# # Save frame in .bin file.
bin_file.append(grey_frame)
# # Save frame as .png file.
grey_frame_filename = "grey.png"
grey_frame_path = os.path.join(frames_path, grey_frame_filename)
save_frame(grey_frame_path, grey_frame)
# Save reference frames.
for reference_index in reference_indices:
# Get reference frame.
reference_image = load_reference_image(reference_index,
reference_images_path)
reference_frame = float_frame_to_uint8_frame(reference_image)
# Save frame in .bin file.
bin_file.append(reference_frame)
# Save frame as .png file.
reference_frame_filename = "reference_{:05d}.png".format(
reference_index)
reference_frame_path = os.path.join(frames_path,
reference_frame_filename)
save_frame(reference_frame_path, reference_frame)
# Save perturbed frames.
for reference_index in tqdm.tqdm(reference_indices):
reference_image = load_reference_image(reference_index,
reference_images_path)
for perturbation_pattern_index in perturbation_patterns_indices:
perturbation_pattern = load_perturbation_pattern(
perturbation_pattern_index, perturbation_patterns_path)
for perturbation_amplitude_index in perturbation_amplitudes_indices:
# Get perturbed frame.
perturbation_amplitude = perturbation_amplitudes[
perturbation_amplitude_index]
perturbed_frame = get_perturbed_frame(reference_image,
perturbation_pattern,
perturbation_amplitude,
config)
perturbed_frame = float_frame_to_uint8_frame(perturbed_frame)
# Save frame in .bin file.
bin_file.append(perturbed_frame)
# Save frame as .png file.
perturbed_frame_filename = "perturbed_r{:05d}_p{:05d}_a{:05d}.png".format(
reference_index, perturbation_pattern_index,
perturbation_amplitude_index)
perturbed_frame_path = os.path.join(frames_path,
perturbed_frame_filename)
save_frame(perturbed_frame_path, perturbed_frame)
# TODO check the following lines!
# Save randomly perturbed frames (if necessary).
if with_random_patterns:
for reference_index in tqdm.tqdm(reference_indices):
reference_image = load_reference_image(reference_index,
reference_images_path)
for perturbation_pattern_index in random_patterns_indices:
pattern = load_perturbation_pattern(
perturbation_pattern_index, perturbation_patterns_path)
# Get perturbed frame.
amplitude = float(15) / float(256) # TODO change this value?
frame = get_perturbed_frame(reference_image, pattern,
amplitude, config)
frame = float_frame_to_uint8_frame(frame)
# Save frame in .bin file.
bin_file.append(frame)
# Save frame as .png file (if necessary).
if perturbation_pattern_index < 100:
perturbed_frame_filename = "perturbed_r{:05d}_p{:05d}.png".format(
reference_index, perturbation_pattern_index)
perturbed_frame_path = os.path.join(
frames_path, perturbed_frame_filename)
save_frame(perturbed_frame_path, frame)
bin_file.close()
print("End creating .bin file.")
print("Start creating .vec and .csv files...")
# Create .vec and .csv files.
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(path, vec_filename)
vec_file = open_vec_file(vec_path, nb_displays=nb_displays)
csv_filename = "{}.csv".format(name)
csv_path = os.path.join(path, csv_filename)
csv_file = open_csv_file(csv_path,
columns=['k_min', 'k_max', 'combination_id'])
# Append adaptation.
grey_frame_id = 0
for _ in range(0, nb_frame_displays):
vec_file.append(grey_frame_id)
# For each repetition...
for repetition_index in range(0, nb_repetitions):
# Add frozen patterns.
combination_indices = permutations[repetition_index]
for combination_index in combination_indices:
combination_frame_id = combination_index
k_min = vec_file.get_display_index() + 1
# Append trial.
for _ in range(0, nb_frame_displays):
vec_file.append(combination_frame_id)
k_max = vec_file.get_display_index()
csv_file.append(k_min=k_min,
k_max=k_max,
combination_id=combination_index)
# Append intertrial.
for _ in range(0, nb_frame_displays):
vec_file.append(grey_frame_id)
# TODO add a random pattern.
# Add random patterns (if necessary).
if with_random_patterns:
random_combination_indices = random_combination_groups[
repetition_index]
for combination_index in random_combination_indices:
combination_frame_id = combination_index
k_min = vec_file.get_display_index() + 1
# Append trial.
for _ in range(0, nb_frame_displays):
vec_file.append(combination_frame_id)
k_max = vec_file.get_display_index()
csv_file.append(k_min=k_min,
k_max=k_max,
combination_id=combination_index)
# Append intertrial.
for _ in range(0, nb_frame_displays):
vec_file.append(grey_frame_id)
csv_file.close()
vec_file.close()
print("End creating .vec and .csv files.")
return
def generate(args):
config = handle_arguments_and_configurations(name, args)
pixel_size = 3.5 # µm
# dmd_width = 1920 # px
# dmd_height = 1080 # px
frame_width_in_um = config['frame']['width']
frame_height_in_um = config['frame']['height']
frame_rate = config['frame']['rate']
nb_repetitions = config['nb_repetitions']
path = config['path']
if not os.path.isdir(path):
os.makedirs(path)
print(path)
dtype = np.uint8
nb_grey_levels = np.iinfo(dtype).max - np.iinfo(dtype).min + 1
nb_images = nb_grey_levels
frame_height_in_px, frame_width_in_px = shape(pixel_size,
width=frame_width_in_um,
height=frame_height_in_um)
# Create .bin file.
bin_filename = "{}.bin".format(name)
bin_path = os.path.join(path, bin_filename)
bin_file = open_bin_file(bin_path,
nb_images,
frame_width=frame_width_in_px,
frame_height=frame_height_in_px,
mode='w')
for k in range(0, nb_grey_levels):
grey_level = float(k) / nb_grey_levels
frame = get_grey_frame(frame_width_in_px,
frame_height_in_px,
luminance=grey_level)
frame = float_frame_to_uint8_frame(frame)
bin_file.append(frame)
bin_file.flush()
bin_file.close()
pattern_config = config['pattern']
pattern = get_pattern(pattern_config, frame_rate)
intertrial_duration = config['intertrial_duration']
initial_adaptation_duration = config['initial_adaptation_duration']
assert (intertrial_duration * frame_rate).is_integer()
# Create .csv file for pattern profile.
csv_filename = "{}_luminance_profile.csv".format(name)
csv_path = os.path.join(path, csv_filename)
columns = ['luminance']
csv_file = open_csv_file(csv_path, columns=columns)
luminances = digitize(pattern)
for luminance in luminances:
csv_file.append(luminance=luminance)
csv_file.close()
# Plot pattern profile.
plot_filename = "{}.pdf".format(name)
plot_path = os.path.join(path, plot_filename)
fig, ax = plot_pattern(pattern, frame_rate)
fig.savefig(plot_path)
plt.close(fig)
# nb_displays_per_trial = int(np.round(trial_duration * frame_rate))
nb_displays_per_trial = pattern.size
nb_displays_per_intertrial = int(np.round(intertrial_duration *
frame_rate))
nb_displays_in_initial_adaptation = int(
np.round(initial_adaptation_duration * frame_rate))
nb_trials = nb_repetitions
nb_intertrials = nb_trials
nb_displays = \
nb_displays_in_initial_adaptation \
+ nb_trials * nb_displays_per_trial \
+ nb_intertrials * nb_displays_per_intertrial
frame_indices = digitize(pattern)
# Create .vec and .csv files.
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(path, vec_filename)
vec_file = open_vec_file(vec_path, nb_displays=nb_displays)
csv_filename = "{}.csv".format(name)
csv_path = os.path.join(path, csv_filename)
csv_file = open_csv_file(csv_path,
columns=['k_min', 'k_max', 'combination_id'])
# Append initial adaptation.
for _ in range(0, nb_displays_in_initial_adaptation):
frame_index = 0
vec_file.append(frame_index)
# For each repetition...
for _ in range(0, nb_repetitions):
k_min = vec_file.get_display_index() + 1
# Append trial.
for k in range(0, nb_displays_per_trial):
frame_id = frame_indices[k]
vec_file.append(frame_id)
k_max = vec_file.get_display_index()
csv_file.append(k_min=k_min, k_max=k_max, combination_id=0)
# Append intertrial.
for _ in range(0, nb_displays_per_intertrial):
frame_id = 0
vec_file.append(frame_id)
csv_file.close()
vec_file.close()
# Print the total duration.
duration = float(nb_displays) / frame_rate
print("duration: {} s".format(duration))
print("duration: {} min".format(duration / 60.0))
return
def generate(args):
config = handle_arguments_and_configurations(name, args)
base_path = config['path']
if not os.path.isdir(base_path):
os.makedirs(base_path)
print("Generation in {}.".format(base_path))
# Get configuration parameters.
stimuli_dirnames = config['stimuli']
mean_luminance = config['mean_luminance']
display_rate = config['display_rate']
adaptation_duration = config['adaptation_duration']
flash_duration = config['flash_duration']
inter_flash_duration = config['inter_flash_duration']
frame_width = config['frame']['width']
frame_height = config['frame']['height']
# nb_repetitions = config['nb_repetitions'] # TODO remove?
seed = config['seed']
# ...
stimulus_nbs = []
stimuli_params = {}
for stimulus_nb, stimulus_dirname in stimuli_dirnames.items():
stimulus_nb = int(stimulus_nb) # TODO improve?
assert os.path.isdir(stimulus_dirname), stimulus_dirname
stimulus_nbs.append(stimulus_nb)
stimuli_params[stimulus_nb] = {
'dirname': stimulus_dirname,
'name': os.path.split(stimulus_dirname)[-1],
}
stimulus_nbs = np.array(stimulus_nbs)
# Set conditions parameters.
condition_nb = 0
condition_params = collections.OrderedDict()
for stimulus_nb in stimulus_nbs:
assert condition_nb not in condition_params
condition_params[condition_nb] = collections.OrderedDict([
('stimulus_nb', stimulus_nb)
])
condition_nb += 1
condition_nbs = np.array(list(condition_params))
nb_conditions = len(condition_nbs)
_ = nb_conditions
# ...
stimuli_nb_trials = {}
stimuli_condition_nbs = {} # stimulus_nb, trial_nb -> condition_nb
stimuli_bin_frame_nbs = { # stimulus_nb, condition_nb -> bin_frame_nb
None: 0, # i.e. inter-flash frame (grey frame)
}
for stimulus_nb in stimulus_nbs:
stimulus_params = stimuli_params[stimulus_nb]
# ...
stimulus_trial_csv_dirname = stimulus_params['dirname']
stimulus_trial_csv_filename = '{}_trials.csv'.format(stimulus_params['name'])
stimulus_trial_csv_path = os.path.join(stimulus_trial_csv_dirname, stimulus_trial_csv_filename)
stimulus_trials = load_csv_file(stimulus_trial_csv_path)
# ...
stimulus_condition_nbs = {} # trial_nb -> condition_nb
stimulus_start_frame_nbs = {} # condition_nb -> start_frame_nb
for trial_nb, stimulus_trial in stimulus_trials.iterrows():
stimulus_condition_nbs[trial_nb] = stimulus_trial['condition_nb']
stimulus_start_frame_nbs[trial_nb] = stimulus_trial['start_frame_nb']
# ...
stimulus_vec_dirname = stimulus_params['dirname']
stimulus_vec_filename = '{}.vec'.format(stimulus_params['name'])
stimulus_vec_path = os.path.join(stimulus_vec_dirname, stimulus_vec_filename)
stimulus_vec = load_vec_file(stimulus_vec_path)
# ...
stimulus_bin_frame_nbs = {}
nb_trials = len(stimulus_trials)
for trial_nb in range(0, nb_trials):
condition_nb = stimulus_condition_nbs[trial_nb]
start_frame_nb = stimulus_start_frame_nbs[condition_nb]
if condition_nb not in stimulus_bin_frame_nbs:
stimulus_bin_frame_nbs[condition_nb] = stimulus_vec[start_frame_nb]
else:
assert stimulus_bin_frame_nbs[condition_nb] == stimulus_vec[start_frame_nb]
# ...
stimuli_nb_trials[stimulus_nb] = len(stimulus_trials)
stimuli_condition_nbs[stimulus_nb] = stimulus_condition_nbs
stimuli_bin_frame_nbs[stimulus_nb] = stimulus_bin_frame_nbs
stimulus_sequence = np.concatenate(tuple([
np.repeat(stimulus_nb, len(stimuli_condition_nbs[stimulus_nb]))
for stimulus_nb in stimulus_nbs
]))
np.random.seed(seed)
np.random.shuffle(stimulus_sequence)
stimuli_indices = {
stimulus_nb: np.where(stimulus_sequence == stimulus_nb)[0]
for stimulus_nb in stimulus_nbs
}
trials = {} # trial_nb -> stimulus_nb, condition_nb
trial_nb = 0
ordering = np.empty_like(stimulus_sequence, dtype=np.int)
for stimulus_nb in stimulus_nbs:
stimulus_condition_nbs = stimuli_condition_nbs[stimulus_nb]
stimulus_indices = stimuli_indices[stimulus_nb]
for condition_nb, stimulus_index in zip(stimulus_condition_nbs.values(), stimulus_indices):
trials[trial_nb] = (stimulus_nb, condition_nb)
ordering[stimulus_index] = trial_nb
trial_nb += 1
nb_trials = len(trials)
# # Set ordering.
# np.random.seed(seed)
# trial_nbs = np.arange(0, nb_trials)
# ordering = np.copy(trial_nbs)
# np.random.shuffle(ordering)
# Create conditions .csv file.
# TODO complete.
# Get number of images in .bin files.
stimuli_nb_bin_images = {}
for stimulus_nb in stimulus_nbs:
stimulus_params = stimuli_params[stimulus_nb]
stimulus_bin_dirname = stimulus_params['dirname']
stimulus_bin_filename = '{}.bin'.format(stimulus_params['name'])
stimulus_bin_path = os.path.join(stimulus_bin_dirname, stimulus_bin_filename)
stimulus_bin_file = open_bin_file(stimulus_bin_path, mode='r')
stimuli_nb_bin_images[stimulus_nb] = stimulus_bin_file.nb_frames
# TODO Map stimulus bin frame numbers to bin frame numbers.
bin_frame_nbs = {
None: 0,
}
bin_frame_nb_offset = 1
for stimulus_nb in stimulus_nbs:
bin_frame_nbs[stimulus_nb] = {}
stimulus_bin_frame_nbs = stimuli_bin_frame_nbs[stimulus_nb]
for condition_nb in stimulus_bin_frame_nbs.keys():
stimulus_bin_frame_nb = stimulus_bin_frame_nbs[condition_nb]
bin_frame_nbs[stimulus_nb][condition_nb] = stimulus_bin_frame_nb + bin_frame_nb_offset
bin_frame_nb_offset += stimuli_nb_bin_images[stimulus_nb]
# Create .bin file.
bin_filename = '{}.bin'.format(name)
bin_path = os.path.join(base_path, bin_filename)
nb_bin_images = 1 + int(np.sum([n for n in stimuli_nb_bin_images.values()]))
# Open .bin file.
bin_file = open_bin_file(bin_path, nb_bin_images, frame_width=frame_width, frame_height=frame_height, reverse=False, mode='w')
# Add grey frame.
grey_frame = get_grey_frame(frame_width, frame_height, luminance=mean_luminance)
grey_frame = float_frame_to_uint8_frame(grey_frame)
bin_file.append(grey_frame)
# Add frames.
for stimulus_nb in stimulus_nbs:
stimulus_params = stimuli_params[stimulus_nb]
stimulus_bin_dirname = stimulus_params['dirname']
stimulus_bin_filename = '{}.bin'.format(stimulus_params['name'])
stimulus_bin_path = os.path.join(stimulus_bin_dirname, stimulus_bin_filename)
stimulus_bin_file = open_bin_file(stimulus_bin_path, mode='r')
assert stimulus_bin_file.width == frame_width
assert stimulus_bin_file.height == frame_height
frame_nbs = stimulus_bin_file.get_frame_nbs()
for frame_nb in frame_nbs:
frame_bytes = stimulus_bin_file.read_frame_as_bytes(frame_nb)
bin_file.append(frame_bytes)
# Close .bin file.
bin_file.close()
# ...
print("End of .bin file creation.")
# Create .vec file.
print("Start creating .vec file...")
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(base_path, vec_filename)
csv_filename = "{}_trials.csv".format(name)
csv_path = os.path.join(base_path, csv_filename)
# ...
nb_displays_during_adaptation = int(np.ceil(adaptation_duration * display_rate))
nb_displays_per_flash = int(np.ceil(flash_duration * display_rate))
nb_displays_per_inter_flash = int(np.ceil(inter_flash_duration * display_rate))
nb_displays_per_trial = nb_displays_per_flash + nb_displays_per_inter_flash
nb_displays = nb_displays_during_adaptation + nb_trials * nb_displays_per_trial
# Open .vec file.
vec_file = open_vec_file(vec_path, nb_displays=nb_displays)
# Open .csv file.
csv_file = open_csv_file(csv_path, columns=['stimulus_nb', 'start_frame_nb', 'end_frame_nb'])
# Add adaptation.
# TODO swap and clean the 2 following lines.
# bin_frame_nb = stimuli_bin_frame_nbs[None] # i.e. default frame (grey)
bin_frame_nb = bin_frame_nbs[None] # i.e. default frame (grey)
for _ in range(0, nb_displays_during_adaptation):
vec_file.append(bin_frame_nb)
# TODO remove the following commented lines.
# # Add repetitions.
# for repetition_nb in tqdm.tqdm(range(0, nb_repetitions)):
# condition_nbs = repetition_orderings[repetition_nb]
# for condition_nb in condition_nbs:
# # Add flash.
# start_frame_nb = vec_file.get_display_nb() + 1
# bin_frame_nb = bin_frame_nbs[condition_nb]
# for _ in range(0, nb_displays_per_flash):
# vec_file.append(bin_frame_nb)
# end_frame_nb = vec_file.get_display_nb()
# csv_file.append(condition_nb=condition_nb, start_frame_nb=start_frame_nb, end_frame_nb=end_frame_nb)
# # Add inter flash.
# bin_frame_nb = bin_frame_nbs[None] # i.e. default frame (grey)
# for _ in range(0, nb_displays_per_inter_flash):
# vec_file.append(bin_frame_nb)
for trial_nb in tqdm.tqdm(ordering):
stimulus_nb, condition_nb = trials[trial_nb]
# Add flash.
start_frame_nb = vec_file.get_display_nb() + 1
# TODO swap and clean the 2 following lines.
# bin_frame_nb = stimuli_bin_frame_nbs[stimulus_nb][condition_nb]
bin_frame_nb = bin_frame_nbs[stimulus_nb][condition_nb]
for _ in range(0, nb_displays_per_flash):
vec_file.append(bin_frame_nb)
end_frame_nb = vec_file.get_display_nb()
csv_file.append(stimulus_nb=stimulus_nb, start_frame_nb=start_frame_nb, end_frame_nb=end_frame_nb)
# Add inter flash.
# TODO swap and clean the 2 following lines.
# bin_frame_nb = stimuli_bin_frame_nbs[None]
bin_frame_nb = bin_frame_nbs[None]
for _ in range(0, nb_displays_per_inter_flash):
vec_file.append(bin_frame_nb)
# Close .csv file.
csv_file.close()
# Close .vec file.
vec_file.close()
# ...
print("End of .vec file creation.")
# TODO create conditions .csv file for each stimulus.
return
def generate(args):
config = handle_arguments_and_configurations(name, args)
frame_width_in_um = config['frame']['width']
frame_height_in_um = config['frame']['height']
frame_rate = config['frame']['rate']
frame_resolution = config['frame']['resolution']
background_luminance = config['background_luminance']
size = config['size']
duration = config['duration']
nb_repetitions = config['nb_repetitions']
path = config['path']
pixel_size = frame_resolution * 1e+6 # µm
# Check duration.
assert (duration * frame_rate).is_integer()
assert int(duration * frame_rate) % 2 == 0
# Create output directory (if necessary).
if not os.path.isdir(path):
os.makedirs(path)
print(path)
# Collect frame parameters.
frame_height_in_px, frame_width_in_px = shape(pixel_size,
width=frame_width_in_um,
height=frame_height_in_um)
x, y = linspace(pixel_size,
width=frame_width_in_um,
height=frame_height_in_um)
xm = np.logical_and(-size / 2.0 <= x, x <= +size / 2.0)
ym = np.logical_and(-size / 2.0 <= y, y <= +size / 2.0)
i = np.nonzero(xm)[0]
j = np.nonzero(ym)[0]
i_min, i_max = i[0], i[-1]
j_min, j_max = j[0], j[-1]
# Create white frame.
white_frame = get_grey_frame(frame_width_in_px,
frame_height_in_px,
luminance=background_luminance)
white_frame[j_min:j_max + 1, i_min:i_max + 1] = 1.0
white_frame = float_frame_to_uint8_frame(white_frame)
# Create black frame.
black_frame = get_grey_frame(frame_width_in_px,
frame_height_in_px,
luminance=background_luminance)
black_frame[j_min:j_max + 1, i_min:i_max + 1] = 0.0
black_frame = float_frame_to_uint8_frame(black_frame)
nb_images = 2
# Create .bin file.
bin_filename = "{}.bin".format(name)
bin_path = os.path.join(path, bin_filename)
bin_file = open_bin_file(bin_path,
nb_images,
frame_width=frame_width_in_px,
frame_height=frame_height_in_px,
mode='w')
bin_file.append(black_frame)
bin_file.append(white_frame)
bin_file.close()
nb_displays_per_repetition = int(duration * frame_rate)
nb_displays = nb_displays_per_repetition * nb_repetitions
# Create .vec file.
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(path, vec_filename)
vec_file = open_vec_file(vec_path, nb_displays)
for _ in range(0, nb_repetitions):
for _ in range(0, nb_displays_per_repetition // 2):
frame_index = 1 # i.e.white
vec_file.append(frame_index)
for _ in range(0, nb_displays_per_repetition // 2):
frame_index = 0 # i.e. black
vec_file.append(frame_index)
vec_file.close()
return
def generate(args):
config = handle_arguments_and_configurations(name, args)
base_path = config['path']
if not os.path.isdir(base_path):
os.makedirs(base_path)
print("Generation in {}.".format(base_path))
# Get configuration parameters.
stimulus_names = config['stimuli']
mean_luminance = config['mean_luminance']
display_rate = config['display_rate']
adaptation_duration = config['adaptation_duration']
flash_duration = config['flash_duration']
inter_flash_duration = config['inter_flash_duration']
frame_width = config['frame']['width']
frame_height = config['frame']['height']
seed = config['seed']
to_interleave = config['to_interleave']
# TODO pour chaque stimulus, récupérer les condition_nbs.
# TODO pour chaque stimulus, récupérer le mapping condition_nb -> bin_frame_nb.
# TODO pour chaque stimulus, récupérer la séquence de condition_nb.
# TODO créer une séquence 'interleaved' de (stimulus_nb, condition_nb).
# TODO convertir en séquence (stimulus_nb, bin_frame_nb).
# TODO convertir en séquence bin_frame_nb.
nb_stimuli = len(stimulus_names)
stimulus_nbs = np.arange(0, nb_stimuli)
stimuli_params_dict = {} # stimulus_nb -> stimulus_params
for stimulus_nb in stimulus_nbs:
stimulus_name = stimulus_names[stimulus_nb]
stimulus_params = {
'name':
stimulus_name,
'bin_path':
os.path.join(pystim_path, stimulus_name,
'{}.bin'.format(stimulus_name)),
'vec_path':
os.path.join(pystim_path, stimulus_name,
'{}.vec'.format(stimulus_name)),
'trials_path':
os.path.join(pystim_path, stimulus_name,
'{}_trials.csv'.format(stimulus_name)),
}
stimuli_params_dict[stimulus_nb] = stimulus_params
# Get number of bin frames for each stimulus.
for stimulus_nb in stimulus_nbs:
stimulus_bin_path = stimuli_params_dict[stimulus_nb]['bin_path']
stimulus_nb_bin_frames = load_nb_bin_frames(stimulus_bin_path)
stimuli_params_dict[stimulus_nb][
'nb_bin_frames'] = stimulus_nb_bin_frames
# Create .bin file.
bin_filename = '{}.bin'.format(name)
bin_path = os.path.join(base_path, bin_filename)
nb_bin_frames = 1 + int(
np.sum([
stimuli_params_dict[stimulus_nb]['nb_bin_frames']
for stimulus_nb in stimulus_nbs
]))
# Open .bin file.
bin_file = open_bin_file(bin_path,
nb_bin_frames,
frame_width=frame_width,
frame_height=frame_height,
reverse=False,
mode='w')
# ...
bin_frame_nbs_dict = {
} # stimulus_nb -> stimulus_bin_frame_nb -> bin_frame_nb
# Add grey frame.
grey_frame = get_grey_frame(frame_width,
frame_height,
luminance=mean_luminance)
grey_frame = float_frame_to_uint8_frame(grey_frame)
bin_file.append(grey_frame)
bin_frame_nbs_dict[None] = bin_file.get_frame_nb()
# Add frames.
for stimulus_nb in stimulus_nbs:
stimulus_params = stimuli_params_dict[stimulus_nb]
# Open stimulus .bin file.
stimulus_bin_path = stimulus_params['bin_path']
stimulus_bin_file = open_bin_file(stimulus_bin_path, mode='r')
# Copy frames from stimulus .bin file to .bin file.
stimulus_bin_frame_nbs = stimulus_bin_file.get_frame_nbs()
bin_frame_nbs_dict[stimulus_nb] = {}
for stimulus_bin_frame_nb in stimulus_bin_frame_nbs:
frame_bytes = stimulus_bin_file.read_frame_as_bytes(
stimulus_bin_frame_nb)
bin_file.append(frame_bytes)
bin_frame_nbs_dict[stimulus_nb][
stimulus_bin_frame_nb] = bin_file.get_frame_nb()
# Close stimulus .bin file.
stimulus_bin_file.close()
# ...
assert bin_file.get_frame_nb() == nb_bin_frames - 1, "{} != {} - 1".format(
bin_file.get_frame_nb(), nb_bin_frames)
# Close .bin file.
bin_file.close()
# ...
print("End of .bin file creation.")
# Get trials for each stimulus.
for stimulus_nb in stimulus_nbs:
stimulus_trials_path = stimuli_params_dict[stimulus_nb]['trials_path']
stimulus_trials = load_csv_file(stimulus_trials_path,
expected_columns=[
'condition_nb', 'start_display_nb',
'end_display_nb'
])
stimuli_params_dict[stimulus_nb]['trials'] = stimulus_trials
# Compute the number of trials for each stimulus.
for stimulus_nb in stimulus_nbs:
stimulus_trials = stimuli_params_dict[stimulus_nb]['trials']
stimulus_nb_trials = len(stimulus_trials)
stimuli_params_dict[stimulus_nb]['nb_trials'] = stimulus_nb_trials
# Compute the number of trials after the merger.
nb_trials = int(
np.sum([
stimuli_params_dict[stimulus_nb]['nb_trials']
for stimulus_nb in stimulus_nbs
]))
# Generate the interleaved sequence of stimulus numbers.
merged_stimulus_nbs_sequence = []
for stimulus_nb in stimulus_nbs:
stimulus_nb_trials = stimuli_params_dict[stimulus_nb]['nb_trials']
merged_stimulus_nbs_sequence.extend(stimulus_nb_trials * [stimulus_nb])
merged_stimulus_nbs_sequence = np.array(merged_stimulus_nbs_sequence)
assert merged_stimulus_nbs_sequence.size == nb_trials
if to_interleave:
np.random.seed(seed)
np.random.shuffle(merged_stimulus_nbs_sequence)
# Compute the corresponding interleaved sequence of stimulus condition numbers.
merged_stimulus_condition_nbs_sequence = np.empty(nb_trials, dtype=np.int)
for stimulus_nb in stimulus_nbs:
indices = np.where(merged_stimulus_nbs_sequence == stimulus_nb)[0]
merged_stimulus_condition_nbs_sequence[indices] = stimuli_params_dict[
stimulus_nb]['trials']['condition_nb']
# Compute the corresponding interleaved sequence of stimulus bin frame numbers.
merged_stimulus_bin_frame_nbs_sequence = np.empty(nb_trials, dtype=np.int)
for stimulus_nb in stimulus_nbs:
stimulus_trials = stimuli_params_dict[stimulus_nb]['trials']
stimulus_vec_path = stimuli_params_dict[stimulus_nb]['vec_path']
stimulus_bin_frame_nbs_sequence = load_vec_file(stimulus_vec_path)
tmp = np.empty(len(stimulus_trials), dtype=np.int)
for index, stimulus_trial in stimulus_trials.iterrows():
start_display_nb = stimulus_trials['start_display_nb'][index]
end_display_nb = stimulus_trials['end_display_nb'][index]
try:
bin_frame_nbs_sequence = stimulus_bin_frame_nbs_sequence[
start_display_nb:end_display_nb + 1]
except TypeError as e:
print(start_display_nb)
print(end_display_nb)
raise e
assert np.all(bin_frame_nbs_sequence == bin_frame_nbs_sequence[0])
bin_frame_nb = bin_frame_nbs_sequence[0]
tmp[index] = bin_frame_nb
indices = np.where(merged_stimulus_nbs_sequence == stimulus_nb)[0]
merged_stimulus_bin_frame_nbs_sequence[indices] = tmp
# Summarize everything in a data frame.
merged_trials_dataframe = pd.DataFrame(
data={
'stimulus_nb': merged_stimulus_nbs_sequence,
'condition_nb': merged_stimulus_condition_nbs_sequence,
'bin_frame_nb': merged_stimulus_bin_frame_nbs_sequence,
})
# Create .vec file.
print("Start creating .vec file...")
vec_filename = "{}.vec".format(name)
vec_path = os.path.join(base_path, vec_filename)
csv_filename = "{}_trials.csv".format(name)
csv_path = os.path.join(base_path, csv_filename)
csv_filenames = {
stimulus_nb: os.path.join(
base_path, '{}_{}_trials.csv'.format(name,
stimulus_names[stimulus_nb]))
for stimulus_nb in stimulus_nbs
}
csv_paths = {
stimulus_nb: os.path.join(base_path, csv_filenames[stimulus_nb])
for stimulus_nb in stimulus_nbs
}
# ...
nb_displays_during_adaptation = int(
np.ceil(adaptation_duration * display_rate))
nb_displays_per_flash = int(np.ceil(flash_duration * display_rate))
nb_displays_per_inter_flash = int(
np.ceil(inter_flash_duration * display_rate))
nb_displays_per_trial = nb_displays_per_flash + nb_displays_per_inter_flash
nb_displays = nb_displays_during_adaptation + nb_trials * nb_displays_per_trial
# # Open .vec file.
vec_file = open_vec_file(vec_path, nb_displays=nb_displays)
# # Open .csv files.
columns = ['condition_nb', 'start_display_nb', 'end_display_nb']
csv_file = open_csv_file(csv_path, columns=columns)
csv_files = {
stimulus_nb: open_csv_file(csv_paths[stimulus_nb], columns=columns)
for stimulus_nb in stimulus_nbs
}
# # Add adaptation.
bin_frame_nb = bin_frame_nbs_dict[None] # i.e. default frame (grey)
for _ in range(0, nb_displays_during_adaptation):
vec_file.append(bin_frame_nb)
for _, merged_trial in tqdm.tqdm(merged_trials_dataframe.iterrows()):
stimulus_nb = merged_trial['stimulus_nb']
stimulus_condition_nb = merged_trial['condition_nb']
stimulus_bin_frame_nb = merged_trial['bin_frame_nb']
# Add flash.
start_display_nb = vec_file.get_display_nb() + 1
bin_frame_nb = bin_frame_nbs_dict[stimulus_nb][stimulus_bin_frame_nb]
for _ in range(0, nb_displays_per_flash):
vec_file.append(bin_frame_nb)
end_display_nb = vec_file.get_display_nb()
csv_file.append(condition_nb=stimulus_nb,
start_display_nb=start_display_nb,
end_display_nb=end_display_nb)
csv_files[stimulus_nb].append(condition_nb=stimulus_condition_nb,
start_display_nb=start_display_nb,
end_display_nb=end_display_nb)
# Add inter flash.
bin_frame_nb = bin_frame_nbs_dict[None]
for _ in range(0, nb_displays_per_inter_flash):
vec_file.append(bin_frame_nb)
# Close .csv files.
csv_file.close()
for csv_file in csv_files.values():
csv_file.close()
# Close .vec file.
vec_file.close()
# ...
print("End of .vec file creation.")
return
