def load_hues(self, arxiv_id: ArxivId,
iteration: str) -> List[HueSearchRegion]:
hues_path = os.path.join(
directories.iteration(
f"sources-with-colorized-{entity_name}",
arxiv_id,
iteration,
),
"entity_hues.csv",
)
if not os.path.exists(hues_path):
logging.warning("Could not find any hues at %s", hues_path)
return []
searches = []
for record in file_utils.load_from_csv(hues_path,
ColorizationRecord):
searches.append(
HueSearchRegion(
hue=record.hue,
record=record,
relative_file_path=None,
masks=None,
))
return searches
def save(self, item: PageRasterPair, result: np.ndarray) -> None:
output_dir = directories.iteration(self.get_output_base_dirkey(),
item.arxiv_id, item.iteration)
image_path = os.path.join(output_dir, item.relative_path,
item.image_name)
image_dir = os.path.dirname(image_path)
if not os.path.exists(image_dir):
os.makedirs(image_dir)
cv2.imwrite(image_path, result)
logging.debug("Diffed images and stored result at %s", image_path)
def load(self) -> Iterator[PageRasterPair]:
for arxiv_id in self.arxiv_ids:
output_dir = directories.arxiv_subdir(
self.get_output_base_dirkey(), arxiv_id)
file_utils.clean_directory(output_dir)
# Get output file names from results of compiling the uncolorized TeX sources.
output_files = get_output_files(
directories.arxiv_subdir("compiled-sources", arxiv_id))
if len(output_files) == 0:
continue
for iteration in directories.iteration_names(
self.get_raster_base_dirkey(), arxiv_id):
original_images_dir = directories.arxiv_subdir(
"paper-images", arxiv_id)
modified_images_dir = directories.iteration(
self.get_raster_base_dirkey(), arxiv_id, iteration)
for output_file in output_files:
relative_file_path = output_file.path
original_images_path = os.path.join(
original_images_dir, relative_file_path)
for img_name in os.listdir(original_images_path):
original_img_path = os.path.join(
original_images_path, img_name)
modified_img_path = os.path.join(
modified_images_dir, relative_file_path, img_name)
if not os.path.exists(modified_img_path):
logging.warning(
"Could not find expected image %s. Skipping diff for this paper.",
modified_img_path,
)
break
original_img = cv2.imread(original_img_path)
modified_img = cv2.imread(modified_img_path)
yield PageRasterPair(
arxiv_id,
iteration,
relative_file_path,
img_name,
original_img,
modified_img,
)
def load(self) -> Iterator[LocationTask]:
for arxiv_id in self.arxiv_ids:
output_dir = directories.arxiv_subdir("citation-cluster-locations",
arxiv_id)
file_utils.clean_directory(output_dir)
boxes_by_hue_iteration = file_utils.load_citation_hue_locations(
arxiv_id)
if boxes_by_hue_iteration is None:
continue
boxes_by_citation_key: Dict[str, List[BoundingBox]] = {}
for iteration in directories.iteration_names(
"sources-with-colorized-citations", arxiv_id):
citation_hues_path = os.path.join(
directories.iteration(
"sources-with-colorized-citations",
arxiv_id,
iteration,
),
"entity_hues.csv",
)
if not os.path.exists(citation_hues_path):
logging.warning(
"Could not find citation hue colors for %s iteration %s. Skipping",
arxiv_id,
iteration,
)
continue
for record in file_utils.load_from_csv(citation_hues_path,
ColorizationRecord):
key = record.entity_id
if key not in boxes_by_citation_key:
boxes_by_citation_key[key] = []
hue_iteration = HueIteration(record.hue, iteration)
boxes_by_citation_key[key].extend(
boxes_by_hue_iteration.get(hue_iteration, []))
for key, boxes in boxes_by_citation_key.items():
yield LocationTask(
arxiv_id=arxiv_id,
citation_key=key,
boxes=boxes,
)
def save(self, item: TexAndTokens, result: ColorizationResult) -> None:
iteration = result.iteration
iteration_id = f"all-files-{iteration}"
output_sources_path = directories.iteration(
"sources-with-colorized-equation-tokens",
item.arxiv_id,
iteration_id,
)
logging.debug("Outputting to %s", output_sources_path)
# Create new directory for each colorization iteration.
unpack_path = unpack(item.arxiv_id, output_sources_path)
sources_unpacked = unpack_path is not None
if unpack_path is None:
logging.warning("Could not unpack sources into %s",
output_sources_path)
if sources_unpacked:
for tex_path, colorized_tex in result.result.colorized_files.items(
):
full_tex_path = os.path.join(output_sources_path, tex_path)
with open(full_tex_path, "w",
encoding=colorized_tex.encoding) as tex_file:
tex_file.write(colorized_tex.contents)
hues_path = os.path.join(output_sources_path, "entity_hues.csv")
for colorized_token in result.result.colorized_tokens:
file_utils.append_to_csv(
hues_path,
EquationTokenColorizationRecord(
entity_id=(str(colorized_token.equation_index) + "-" +
str(colorized_token.token_index)),
hue=colorized_token.hue,
tex_path=colorized_token.tex_path,
iteration=str(iteration),
equation_index=colorized_token.equation_index,
token_index=colorized_token.token_index,
start=colorized_token.start,
end=colorized_token.end,
text=colorized_token.text,
),
)
def save(self, item: ColorizationTask, result: ColorizationResult) -> None:
iteration = result.iteration
colorized_tex = result.tex
colorized_citations = result.colorized_citations
iteration_id = directories.tex_iteration(item.tex_path, str(iteration))
output_sources_path = directories.iteration(
"sources-with-colorized-citations",
item.arxiv_id,
iteration_id,
)
logging.debug("Outputting to %s", output_sources_path)
# Create new directory for each colorization iteration for each TeX file.
unpack_path = unpack(item.arxiv_id, output_sources_path)
sources_unpacked = unpack_path is not None
if unpack_path is None:
logging.warning("Could not unpack sources into %s",
output_sources_path)
if sources_unpacked:
tex_path = os.path.join(output_sources_path, item.tex_path)
with open(tex_path, "w",
encoding=item.file_contents.encoding) as tex_file:
tex_file.write(colorized_tex)
hues_path = os.path.join(output_sources_path, "entity_hues.csv")
# TODO(andrewhead): It might be better to save this CSV data with the same
# encoding as the file the TeX was read from, for the citations, for the
# equations, and for the symbols. There might be some gotchas for character
# positions not lining up between the ones we save using Unicode here and the
# positions in the intended encoding in the original files.
for c in colorized_citations:
record = ColorizationRecord(
hue=c.hue,
entity_id=c.key,
tex_path=item.tex_path,
iteration=iteration_id,
)
file_utils.append_to_csv(hues_path, record)
def save(self, item: ColorizationTask, result: ColorizationResult) -> None:
iteration = result.iteration
colorized_tex = result.tex
entity_hues = result.entity_hues
iteration_id = directories.tex_iteration(item.tex_path, str(iteration))
output_sources_path = directories.iteration(
self.get_output_base_dirkey(),
item.arxiv_id,
iteration_id,
)
logging.debug("Outputting to %s", output_sources_path)
# Each colorization batch gets a new sources directory.
unpack_path = unpack(item.arxiv_id, output_sources_path)
sources_unpacked = unpack_path is not None
if unpack_path is None:
logging.warning("Could not unpack sources into %s",
output_sources_path)
if sources_unpacked:
# Rewrite the TeX with the colorized TeX.
tex_path = os.path.join(output_sources_path, item.tex_path)
with open(tex_path, "w",
encoding=item.file_contents.encoding) as tex_file:
tex_file.write(colorized_tex)
# Save a log of which hues were assigned to which entities.
hues_path = os.path.join(output_sources_path, "entity_hues.csv")
for (hue, entity) in entity_hues:
file_utils.append_to_csv(
hues_path,
ColorizationRecord(
tex_path=item.tex_path,
iteration=str(iteration),
hue=hue,
entity_id=entity.id_,
),
)
def process(self, item: LocationTask) -> Iterator[HueLocationInfo]:
# Filter out entities that are empty (i.e., have nothing to color)
# A '-1' in the 'start' or 'end' field indicates that the entity does not occur in a
# specific place in the TeX, but rather a custom coloring technique based on other
# entity properties will be used. So entities that have a '-1' for their start and
# end should still be processed even though they appear to be zero-length.
entities_filtered = [
e for e in item.entities
if e.start == -1 or e.end == -1 or e.start != e.end
]
# Sort entities by the order in which they appear in the TeX. This allows the pipeline
# to keep track of which ones appear first, when trying to recover from errors (i.e., when
# trying to detect which entity in a batch may have shifted to cause many others to move.)
entities_ordered = sorted(entities_filtered, key=lambda e: e.start)
# Construct a queue of entities to detect.
entities_by_id = {e.id_: e for e in entities_ordered}
to_process = deque([e.id_ for e in entities_ordered])
to_process_alone: Deque[str] = deque()
# Path to output directories. These directories will be redefined once for each batch.
colorized_tex_dir: Optional[str] = None
compiled_tex_dir: Optional[str] = None
raster_output_dir: Optional[str] = None
diffs_output_dir: Optional[str] = None
# Iteration state
batch_index = -1
iteration_id = None
def next_batch() -> List[str]:
"""
Get the next batch of entities to process. First tries to sample a batch from
'to_process', and then attempts to sample individual entities from 'to_process_alone'.
"""
if len(to_process) > 0:
return [
to_process.popleft()
for _ in range(min(self.args.batch_size, len(to_process)))
]
return [to_process_alone.popleft()]
def _cleanup_from_last_batch() -> None:
" Clean up output directories from the last batch. "
if batch_index > -1 and not self.args.keep_intermediate_files:
logging.debug( # pylint: disable=logging-not-lazy
"Deleting intermediate files used to locate entities (i.e., colorized "
+
"sources, compilation results, and rasters) for paper %s iteration %s",
item.arxiv_id,
iteration_id or "''",
)
intermediate_files_dirs = [
colorized_tex_dir,
compiled_tex_dir,
raster_output_dir,
diffs_output_dir,
]
for dir_ in intermediate_files_dirs:
if dir_ and os.path.exists(dir_):
file_utils.clean_directory(dir_)
os.rmdir(dir_)
while len(to_process) > 0 or len(to_process_alone) > 0:
if batch_index > -1:
_cleanup_from_last_batch()
batch_index += 1
logging.debug(
"Locating bounding boxes for batch %d-%d of entities of type %s for paper %s.",
item.group,
batch_index,
self.get_entity_name(),
item.arxiv_id,
)
iteration_id = directories.tex_iteration(
item.tex_path, f"{item.group}-{batch_index}")
# Define output directory locations for this batch.
colorized_tex_dir = directories.iteration(
self.output_base_dirs["sources"], item.arxiv_id, iteration_id)
compiled_tex_dir = directories.iteration(
self.output_base_dirs["compiled-sources"],
item.arxiv_id,
iteration_id,
)
raster_output_dir = directories.iteration(
self.output_base_dirs["paper-images"], item.arxiv_id,
iteration_id)
diffs_output_dir = directories.iteration(
self.output_base_dirs["diffed-images"], item.arxiv_id,
iteration_id)
# Fetch the next batch of entities to process.
batch = next_batch()
entities: List[SerializableEntity] = [
entities_by_id[id_] for id_ in batch
]
# Colorize the TeX for all the entities.
custom_colorize_func = self.get_colorize_func()
logging.debug(
"Attempting to colorize entities in TeX for entity batch %d-%d of paper %s.",
item.group,
batch_index,
item.arxiv_id,
)
if custom_colorize_func is not None:
colorized_tex = custom_colorize_func(
item.file_contents.contents, entities,
self.get_colorize_options())
if len(colorized_tex.entity_hues) == 0:
logging.info( # pylint: disable=logging-not-lazy
"Custom colorization function colored nothing for entity batch %d-%d of "
+
"paper %s when coloring file %s. The function probably decide there was "
+
"nothing to do for this file, and will hopefullly colorize these "
+
"entities in another file. Skipping this batch for this file.",
item.group,
batch_index,
item.arxiv_id,
item.file_contents.path,
)
continue
else:
colorized_tex = colorize_entities(item.file_contents.contents,
entities,
self.get_colorize_options())
# If some entities were skipped during colorization, perhaps because they
# overlapped with each other, add them back to the work queue.
if colorized_tex.skipped is not None and len(
colorized_tex.skipped) > 0:
logging.info( # pylint: disable=logging-not-lazy
"Entities %s were skipped during colorization batch %d-%d for paper "
+ "%s. They will be processed in a later batch.",
[e.id_ for e in colorized_tex.skipped],
item.group,
batch_index,
item.arxiv_id,
)
# Queue skipped entities in the order that they initially appeared in the batch.
reprocess_ids = {e.id_ for e in colorized_tex.skipped}
reprocess_sorted = [
id_ for id_ in batch if id_ in reprocess_ids
]
to_process.extendleft(reversed(reprocess_sorted))
# Remove skipped entities from the current batch.
for skip in colorized_tex.skipped:
del batch[batch.index(skip.id_)]
# Save the colorized TeX to the file system.
save_success = save_colorized_tex(
item.arxiv_id,
colorized_tex_dir,
item.tex_path,
iteration_id,
colorized_tex.tex,
item.file_contents.encoding,
colorized_tex.entity_hues,
)
logging.debug(
"Finished attempting to colorize entities for entity batch %d-%d of paper %s.",
item.group,
batch_index,
item.arxiv_id,
)
if not save_success:
logging.error( # pylint: disable=logging-not-lazy
"Failed to save colorized TeX files for arXiv paper %s. "
"This paper will be skipped.",
item.arxiv_id,
)
# Compile the TeX with the colors.
shutil.copytree(colorized_tex_dir, compiled_tex_dir)
compilation_result = compile_tex(compiled_tex_dir)
save_compilation_result("compiled-sources", item.arxiv_id,
compiled_tex_dir, compilation_result)
if not compilation_result.success:
# If colorizing a specific entity caused the failure, remove the entity that caused
# the problem from the batch and restart with a new batch, minus this entity.
last_colorized_entity_id = get_last_colorized_entity(
item.arxiv_id, compiled_tex_dir)
if last_colorized_entity_id is not None:
problem_ids = [last_colorized_entity_id]
if batch.index(last_colorized_entity_id) < len(batch) - 1:
problem_ids += [
batch[batch.index(last_colorized_entity_id) + 1]
]
if len(batch) == 1:
logging.warning( # pylint: disable=logging-not-lazy
"Failed to compile paper %s with colorized entity %s, even when it was "
+
"colorized in isolation. The location of this entity will not be detected.",
item.arxiv_id,
batch[0],
)
continue
logging.warning( # pylint: disable=logging-not-lazy
"Failed to compile paper %s with colorized entities. The culprit may be "
+
"the colorization command for entity %s. The problematic entities will be "
+
"colorized on their own, and the rest of the entities will be colorized "
+ "together in the next batch.",
item.arxiv_id,
" or ".join(problem_ids),
)
for id_ in problem_ids:
to_process_alone.append(id_)
del batch[batch.index(id_)]
to_process.extendleft(reversed(batch))
continue
# If there was some other reason for the error, remove just the first entity from the batch.
logging.error( # pylint: disable=logging-not-lazy
"Failed to compile paper %s with colorized entities %s. The cause "
+
"is assumed to be in the first colorized entity. The location for the "
+
"first entity %s will not be detected. The remainder of the entities in "
+ "this batch will be processed in another batch.",
item.arxiv_id,
batch,
batch[0],
)
del [batch[0]]
to_process.extendleft(reversed(batch))
continue
# Raster the pages to images, and compute diffs from the original images.
output_files = compilation_result.output_files
for output_file in output_files:
raster_success = raster_pages(
compiled_tex_dir,
os.path.join(raster_output_dir,
directories.escape_slashes(output_file.path)),
output_file.path,
output_file.output_type,
)
if not raster_success:
logging.error( # pylint: disable=logging-not-lazy
"Failed to rasterize pages for %s iteration %s. The locations for entities "
+ "with IDs %s with not be detected.",
item.arxiv_id,
iteration_id,
batch,
)
continue
logging.debug(
"Attempting to diff rastered pages for paper %s iteration %s.",
item.arxiv_id,
iteration_id,
)
diff_success = diff_images_in_raster_dirs(
output_files,
raster_output_dir,
diffs_output_dir,
item.arxiv_id,
)
logging.debug(
"Finished diffing attempt for paper %s iteration %s. Success? %s.",
item.arxiv_id,
iteration_id,
diff_success,
)
if not diff_success:
logging.error( # pylint: disable=logging-not-lazy
"Failed to difference images of original and colorized versions of "
+
"papers %s in batch processing iteration %s. The locations for entities with IDs "
+ "%s will not be detected.",
item.arxiv_id,
iteration_id,
batch,
)
continue
# Locate the entities in the diffed images.
logging.debug(
"Attempting to locate entities using image differences for paper %s iteration %s.",
item.arxiv_id,
iteration_id,
)
entity_hues = colorized_tex.entity_hues
location_result = locate_entities(item.arxiv_id, raster_output_dir,
diffs_output_dir, entity_hues)
if location_result is None:
logging.warning( # pylint: disable=logging-not-lazy
"Error occurred when locating entities by hue in diffed images "
+
"for paper %s. None of the entities in batch %s will be detected.",
item.arxiv_id,
batch,
)
continue
if self.should_sanity_check_images(
) and location_result.black_pixels_found:
logging.warning( # pylint: disable=logging-not-lazy
"Ignoring bounding boxes found for paper %s in batch %s due to "
+
"black pixels found in the images. This might indicate that the colorization "
+ "commands introduced subtle shifts of the text.",
item.arxiv_id,
batch,
)
continue
# If colorizing entities seemed to cause drift in the document...
if len(location_result.shifted_entities) > 0:
logging.warning( # pylint: disable=logging-not-lazy
"Some entities shifted position in the colorized TeX for paper %s batch %s: "
+
"%s. Attempting to remove the first shifted entity from the batch.",
item.arxiv_id,
batch,
location_result.shifted_entities,
)
first_shifted_entity_id = None
for entity_id in batch:
if entity_id in location_result.shifted_entities:
first_shifted_entity_id = entity_id
break
if first_shifted_entity_id is not None:
if len(batch) > 1:
logging.info( # pylint: disable=logging-not-lazy
"Entity %s has been marked as being the potential cause of shifting in "
+
"the colorized document for paper %s batch %d-%d. It will be processed "
+
"later on its own. The other shifted entities in %s will be queued to "
+ "process as a group in an upcoming batch.",
first_shifted_entity_id,
item.arxiv_id,
item.group,
batch_index,
location_result.shifted_entities,
)
# Get the index of the first entity for which the location has shifted
# during colorization.
moved_entity_index = batch.index(
first_shifted_entity_id)
# Mark all other entities that have shifted after the first one one to be processed
# in a later batch (instead of on their own). It could be that they won't shift
# once the first shifted entity is removed.
for i in range(len(batch) - 1, moved_entity_index, -1):
if batch[i] in location_result.shifted_entities:
to_process.appendleft(batch[i])
del batch[i]
# Mark the first entity that shifted to be reprocessed alone, where its position
# might be discoverable, without affecting the positions of other element.
del batch[moved_entity_index]
to_process_alone.append(first_shifted_entity_id)
elif len(batch) == 1 and self.should_sanity_check_images():
logging.info( # pylint: disable=logging-not-lazy
"Skipping entity %s for paper %s as it caused " +
"colorization errors even when colorized in isolation.",
first_shifted_entity_id,
item.arxiv_id,
)
continue
elif len(batch) == 1:
logging.info( # pylint: disable=logging-not-lazy
"Entity %s has been marked as the cause of shifting in "
+
"the colorized document for paper %s. Its location will "
+
"still be saved (if one was found), though this location should be "
+ "considered potentially inaccurate.",
first_shifted_entity_id,
item.arxiv_id,
)
else:
logging.warning( # pylint: disable=logging-not-lazy
"Could not find a single entity that was likely responsible for shifting in "
+
"the colorized version of paper %s batch %d-%d. All entities in batch %s will "
+ "be processed on their own.",
item.arxiv_id,
item.group,
batch_index,
batch,
)
to_process_alone.extend(batch)
logging.debug(
"Finished attempt at locating entities with image diffs for paper %s iteration %s.",
item.arxiv_id,
iteration_id,
)
# The code above is responsible for filtering 'batch' to ensure that it doesn't include
# any entity IDs that shouldn't be saved to file, for example if the client has asked that
# entity IDs that cause colorization errors be omitted from the results.
for entity_id in batch:
for box in location_result.locations[entity_id]:
yield HueLocationInfo(
tex_path=item.tex_path,
iteration=iteration_id,
hue=entity_hues[entity_id],
entity_id=entity_id,
page=box.page,
left=box.left,
top=box.top,
width=box.width,
height=box.height,
)
_cleanup_from_last_batch()
def load(self) -> Iterator[SearchTask]:
for arxiv_id in self.arxiv_ids:
output_dir = directories.arxiv_subdir(
self.get_output_base_dirkey(), arxiv_id)
file_utils.clean_directory(output_dir)
# Get output file names from results of compiling the uncolorized TeX sources.
output_files = get_output_files(
directories.arxiv_subdir("compiled-sources", arxiv_id))
for iteration in directories.iteration_names(
self.get_diff_images_base_dirkey(), arxiv_id):
diff_images_dir = directories.iteration(
self.get_diff_images_base_dirkey(), arxiv_id, iteration)
hue_searches = self.load_hues(arxiv_id, iteration)
hue_searches_by_file: Dict[Path, List[HueSearchRegion]] = {}
for search in hue_searches:
output_paths = [f.path for f in output_files]
files_to_search = ([search.relative_file_path]
if search.relative_file_path is not None
else output_paths)
for path in files_to_search:
if path not in hue_searches_by_file:
hue_searches_by_file[path] = []
hue_searches_by_file[path].append(search)
for relative_file_path, search_regions in hue_searches_by_file.items(
):
diff_images_file_path = os.path.join(
diff_images_dir, relative_file_path)
page_images = {}
colorization_error_detected = False
for img_name in os.listdir(diff_images_file_path):
img_path = os.path.join(diff_images_file_path,
img_name)
page_image = cv2.imread(img_path)
if not self.args.skip_visual_validation:
if contains_black_pixels(page_image):
logging.warning(
"Black pixels found in image diff %s",
img_path)
colorization_error_detected = True
page_number = (int(
os.path.splitext(img_name)[0].replace("page-", ""))
- 1)
page_images[page_number] = page_image
if colorization_error_detected:
logging.warning( # pylint: disable=logging-not-lazy
"Colorization error detected. Skipping hue location for "
+ "iteration %s for arXiv paper %s",
iteration,
arxiv_id,
)
break
for search_region in search_regions:
yield SearchTask(
arxiv_id,
iteration,
page_images,
relative_file_path,
search_region,
)
def load_hues(self, arxiv_id: ArxivId,
iteration: str) -> List[HueSearchRegion]:
equation_boxes_path = os.path.join(
directories.arxiv_subdir("hue-locations-for-equations", arxiv_id),
"hue_locations.csv",
)
bounding_boxes: Dict[EquationId, BoundingBoxesByFile] = {}
for location_info in file_utils.load_from_csv(equation_boxes_path,
HueLocationInfo):
equation_id = EquationId(
tex_path=location_info.tex_path,
equation_index=int(location_info.entity_id),
)
if equation_id not in bounding_boxes:
bounding_boxes[equation_id] = {}
file_path = location_info.relative_file_path
if file_path not in bounding_boxes[equation_id]:
bounding_boxes[equation_id][file_path] = []
box = BoundingBox(
page=location_info.page,
left=location_info.left,
top=location_info.top,
width=location_info.width,
height=location_info.height,
)
bounding_boxes[equation_id][file_path].append(box)
token_records_by_equation: Dict[EquationId, Dict[
int, EquationTokenColorizationRecord]] = {}
token_hues_path = os.path.join(
directories.iteration(
"sources-with-colorized-equation-tokens",
arxiv_id,
iteration,
),
"entity_hues.csv",
)
for record in file_utils.load_from_csv(
token_hues_path, EquationTokenColorizationRecord):
equation_id = EquationId(tex_path=record.tex_path,
equation_index=record.equation_index)
token_index = int(record.token_index)
if equation_id not in token_records_by_equation:
token_records_by_equation[equation_id] = {}
token_records_by_equation[equation_id][token_index] = record
hue_searches = []
for equation_id, boxes_by_file in bounding_boxes.items():
for file_path, boxes in boxes_by_file.items():
masks_by_page: MasksForPages = {}
for box in boxes:
if box.page not in masks_by_page:
masks_by_page[box.page] = []
masks_by_page[box.page].append(
Rectangle(box.left, box.top, box.width, box.height))
if equation_id in token_records_by_equation:
for token_index, record in token_records_by_equation[
equation_id].items():
hue_searches.append(
HueSearchRegion(
hue=record.hue,
record=record,
relative_file_path=file_path,
masks=masks_by_page,
))
return hue_searches