Marmelator/cv-analysis-service
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 37 Wiki Activity

Merge branch 'master' of ssh://git.iqser.com:2222/rr/cv-analysis into fig-detection-scanned-pdfs

Browse Source 
 Conflicts:
	cv_analysis/figure_detection.py
	cv_analysis/layout_parsing.py
	cv_analysis/table_parsing.py
	scripts/annotate.py
This commit is contained in:
llocarnini 2022-05-04 09:33:14 +02:00
commit 0e30e97f80
9 changed files with 99 additions and 466 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Dockerfile
View File

@ -7,6 +7,7 @@ WORKDIR /app/service
COPY ./src ./src
COPY cv_analysis ./cv_analysis
COPY config.yaml ./config.yaml
RUN python3 -m pip install --upgrade pip
RUN python3 -m pip install -e .

2
config.yaml
View File

@ -23,5 +23,5 @@ deskew:
test_dummy: test_dummy
visual_logging:
level: $LOGGING_LEVEL_ROOT|DEBUG
level: $LOGGING_LEVEL_ROOT|INFO
output_folder: /tmp/debug/

2
cv_analysis/figure_detection.py
View File

@ -9,7 +9,7 @@ from cv_analysis.utils.post_processing import remove_included
from cv_analysis.utils.filters import is_large_enough, has_acceptable_format
from cv_analysis.utils.text import remove_primary_text_regions
from cv_analysis.utils.visual_logging import vizlogger
#from PIL import Image
def is_likely_figure(cont, min_area=5000, max_width_to_hight_ratio=6):
return is_large_enough(cont, min_area) and has_acceptable_format(cont, max_width_to_hight_ratio)

1
cv_analysis/redaction_detection.py
View File

@ -51,4 +51,3 @@ def annotate_redactions_in_pdf(pdf_path, page_index=1, show=False):
if show:
show_mpl(page)

14
cv_analysis/table_parsing.py
View File

@ -16,10 +16,8 @@ from cv_analysis.utils.visual_logging import vizlogger
from cv_analysis.layout_parsing import parse_layout
def add_external_contours(image, contour_source_image):
contours, _ = cv2.findContours(contour_source_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
contours = filter(partial(is_large_enough, min_area=5000), contours)
def add_external_contours(image, img):
contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
for cnt in contours:
x, y, w, h = cv2.boundingRect(cnt)
cv2.rectangle(image, (x, y), (x + w, y + h), 255, 1)
@ -80,7 +78,7 @@ def isolate_vertical_and_horizontal_components(img_bin):
img_bin_v = cv2.morphologyEx(img_bin, cv2.MORPH_OPEN, kernel_v)
img_lines_raw = img_bin_v | img_bin_h
vizlogger.debug(img_lines_raw, "tables02_isolate02_img_bin_v.png")
kernel_h = np.ones((1, 30), np.uint8)
kernel_v = np.ones((30, 1), np.uint8)
img_bin_h = cv2.dilate(img_bin_h, kernel_h, iterations=2)
@ -100,7 +98,7 @@ def isolate_vertical_and_horizontal_components(img_bin):
vizlogger.debug(img_bin_final, "tables10_isolate12_threshold.png")
img_bin_final = cv2.dilate(img_bin_final, np.ones((1, 1), np.uint8), iterations=1)
vizlogger.debug(img_bin_final, "tables11_isolate13_dilate.png")
# add contours before lines are extended by blurring
img_bin_final = add_external_contours(img_bin_final, img_lines_raw)
vizlogger.debug(img_bin_final, "tables11_isolate14_contours_added.png")
@ -174,7 +172,7 @@ def parse_table(image: np.array, show=False):
stats = np.vstack(list(filter(is_large_enough, stats)))
rects = stats[:, :-1][2:]
return list(rects)
return list(map(list, rects))
def annotate_tables_in_pdf(pdf_path, page_index=0, deskew=False, show=False):
@ -189,6 +187,8 @@ def annotate_tables_in_pdf(pdf_path, page_index=0, deskew=False, show=False):
stats = parse_table(page)
page = draw_rectangles(page, stats, annotate=True)
vizlogger.debug(page, "tables15_final_output.png")
if show:
show_mpl(page)
def tables_in_image(cropped_image):

460
cv_analysis/test/test_data/table.json
View File

@ -1,406 +1,62 @@
{
"0": [
[
211,
415,
367,
29
],
[
581,
415,
417,
29
],
[
1001,
415,
406,
29
],
[
211,
447,
367,
47
],
[
581,
447,
417,
47
],
[
1001,
447,
406,
47
],
[
211,
497,
367,
47
],
[
580,
497,
418,
47
],
[
1001,
497,
406,
47
],
[
211,
547,
367,
47
],
[
580,
547,
418,
47
],
[
1001,
547,
406,
47
],
[
211,
597,
367,
47
],
[
581,
597,
417,
47
],
[
1001,
597,
406,
48
],
[
212,
647,
366,
48
],
[
581,
647,
417,
48
],
[
1001,
647,
406,
48
],
[
581,
697,
417,
47
],
[
1001,
697,
407,
48
],
[
212,
698,
366,
47
],
[
211,
747,
367,
48
],
[
581,
747,
417,
48
],
[
1001,
748,
407,
47
],
[
211,
798,
367,
47
],
[
581,
798,
417,
47
],
[
1001,
798,
407,
47
],
[
212,
848,
366,
47
],
[
581,
848,
417,
47
],
[
1001,
848,
407,
48
],
[
212,
898,
366,
48
],
[
581,
898,
417,
48
],
[
1001,
898,
407,
48
],
[
212,
949,
366,
33
],
[
581,
949,
827,
33
],
[
462,
1163,
368,
29
],
[
833,
1163,
404,
29
],
[
462,
1195,
368,
48
],
[
833,
1195,
404,
48
],
[
462,
1245,
368,
48
],
[
833,
1245,
404,
47
],
[
462,
1296,
368,
47
],
[
833,
1296,
404,
47
],
[
462,
1346,
368,
47
],
[
833,
1346,
404,
47
],
[
462,
1396,
368,
47
],
[
834,
1396,
403,
47
],
[
462,
1446,
368,
48
],
[
833,
1446,
404,
48
],
[
462,
1496,
368,
48
],
[
833,
1496,
404,
48
],
[
462,
1547,
368,
47
],
[
834,
1547,
403,
47
],
[
462,
1597,
368,
48
],
[
834,
1597,
403,
47
],
[
462,
1647,
368,
48
],
[
833,
1647,
404,
48
],
[
462,
1698,
368,
47
],
[
833,
1698,
404,
47
],
[
462,
1748,
368,
47
],
[
834,
1748,
403,
47
],
[
462,
1798,
368,
47
],
[
834,
1798,
403,
47
],
[
462,
1848,
368,
48
],
[
834,
1848,
403,
48
],
[
462,
1899,
369,
33
],
[
832,
1899,
405,
33
]
[211, 447, 367, 47],
[581, 447, 417, 47],
[1001, 447, 406, 47],
[211, 497, 367, 47],
[580, 497, 418, 47],
[1001, 497, 406, 47],
[211, 547, 367, 47],
[580, 547, 418, 47],
[1001, 547, 406, 47],
[211, 597, 367, 47],
[581, 597, 417, 47],
[1001, 597, 406, 48],
[212, 647, 366, 48],
[581, 647, 417, 48],
[1001, 647, 406, 48],
[581, 697, 417, 47],
[1001, 697, 407, 48],
[212, 698, 366, 47],
[211, 747, 367, 48],
[581, 747, 417, 48],
[1001, 748, 407, 47],
[211, 798, 367, 47],
[581, 798, 417, 47],
[1001, 798, 407, 47],
[212, 848, 366, 47],
[581, 848, 417, 47],
[1001, 848, 407, 48],
[212, 898, 366, 48],
[581, 898, 417, 48],
[1001, 898, 407, 48],
[462, 1195, 368, 48],
[833, 1195, 404, 48],
[462, 1245, 368, 48],
[833, 1245, 404, 47],
[462, 1296, 368, 47],
[833, 1296, 404, 47],
[462, 1346, 368, 47],
[833, 1346, 404, 47],
[462, 1396, 368, 47],
[834, 1396, 403, 47],
[462, 1446, 368, 48],
[833, 1446, 404, 48],
[462, 1496, 368, 48],
[833, 1496, 404, 48],
[462, 1547, 368, 47],
[834, 1547, 403, 47],
[462, 1597, 368, 48],
[834, 1597, 403, 47],
[462, 1647, 368, 48],
[833, 1647, 404, 48],
[462, 1698, 368, 47],
[833, 1698, 404, 47],
[462, 1748, 368, 47],
[834, 1748, 403, 47],
[462, 1798, 368, 47],
[834, 1798, 403, 47],
[462, 1848, 368, 48],
[834, 1848, 403, 48]
]
}

16
cv_analysis/utils/visual_logging.py
View File

@ -2,21 +2,21 @@ import os
from cv_analysis.config import CONFIG
from cv_analysis.utils.display import save_mpl
LEVEL = CONFIG.visual_logging.level
OUTPUT_FOLDER = CONFIG.visual_logging.output_folder
class VisualLogger:
def __init__(self):
self.level_is_debug = LEVEL == "DEBUG"
self.output_folder = OUTPUT_FOLDER
def __init__(self, level, output_folder):
self.level = level
self.output_folder = output_folder
if not os.path.exists(self.output_folder):
os.mkdir(self.output_folder)
def debug(self, img, name):
if self.level_is_debug:
if self.level_is_debug():
output_path = os.path.join(self.output_folder, name)
save_mpl(img, output_path)
def level_is_debug(self):
return self.level == "DEBUG"
vizlogger = VisualLogger()
vizlogger = VisualLogger(CONFIG.visual_logging.level, CONFIG.visual_logging.output_folder)

4
scripts/client_mock.py
View File

@ -34,7 +34,7 @@ def parse_args():
def main(args):
operations = args.operations.split(",")
for operation in operations:
print("****************************")
@ -49,7 +49,7 @@ def main(args):
elif operation == "layout-parsing":
response = requests.post("http://127.0.0.1:5000/layout", data=open(args.pdf_path, "rb"))
else:
raise ValueError("{args.operation} is not a valid value.")
raise ValueError(f"{args.operation} is not a valid value.")
response.raise_for_status()
predictions = response.json()

65
src/run_service.py
View File

@ -8,11 +8,10 @@ from prometheus_flask_exporter import PrometheusMetrics
from waitress import serve
from cv_analysis.utils import npconvert
from cv_analysis.utils.preprocessing import preprocess_pdf_image # TODO
from cv_analysis.table_parsing import parse_table # , detect_tables_in_pdf
from cv_analysis.redaction_detection import find_redactions # , detect_redactions_in_pdf
from cv_analysis.layout_parsing import parse_layout # , detect_layout_in_pdf #TODO
from cv_analysis.figure_detection import detect_figures # , detect_figures_in_pdf #TODO
from cv_analysis.table_parsing import parse_table
from cv_analysis.redaction_detection import find_redactions
from cv_analysis.layout_parsing import parse_layout
from cv_analysis.figure_detection import detect_figures
from cv_analysis.utils.logging import logger
from cv_analysis.utils.preprocessing import open_pdf
from cv_analysis.config import CONFIG
@ -26,7 +25,6 @@ def suppress_user_warnings():
def main():
file_counter = Counter("cv_analysis_file_counter", "count processed files")
# page_counter = Counter("cv_analysis_page_counter", "count pages from processed files")
ram_metric = Gauge("cv_analysis_memory_usage", "Memory usage in Mb")
def start_monitoring():
@ -44,32 +42,28 @@ def main():
@metrics.summary("tables_request_time_seconds", "Time spent processing tables request")
def get_tables():
start_monitoring()
tables = annotate("tables")
# page_counter.inc(npages)
tables = annotate(parse_table)
return tables
@app.route("/redactions", methods=["POST"])
@metrics.summary("redactions_request_time_seconds", "Time spent processing redaction request")
def get_redactions():
start_monitoring()
redactions = annotate("redactions")
# page_counter.inc(npages)
redactions = annotate(find_redactions)
return redactions
@app.route("/figures", methods=["POST"])
@metrics.summary("figures_request_time_seconds", "Time spent processing figures request")
def get_figures():
start_monitoring()
figures = annotate("figures")
# page_counter.inc(npages)
figures = annotate(detect_figures)
return figures
@app.route("/layout", methods=["POST"])
@metrics.summary("layout_request_time_seconds", "Time spent processing layout request")
def get_layout():
start_monitoring()
layout = annotate("layout")
# page_counter.inc(npages)
layout = annotate(parse_layout)
return layout
@app.route("/status", methods=["GET"])
@ -77,7 +71,6 @@ def main():
response = "OK"
return jsonify(response)
# predictor = initialize_predictor()
logger.info("<3 Annotator ready.")
mode = CONFIG.webserver.mode
@ -89,46 +82,30 @@ def main():
tracemalloc.stop()
def apply_annotation_function(annotation_function, page_list):
outdict = {}
for i, page in enumerate(page_list):
results = annotation_function(page)
if results:
outdict.update({i: results})
return outdict
def make_annotations(pdf, task):
if task == "tables":
annotation = {"tables": apply_annotation_function(parse_table, pdf)}
elif task == "redactions":
annotation = {"redactions": apply_annotation_function(find_redactions, pdf)}
elif task == "figures":
annotation = {"figures": apply_annotation_function(detect_figures, pdf)}
elif task == "layout":
annotation = {"layout": apply_annotation_function(parse_layout, pdf)}
else:
raise ValueError(
f"'{task}' is not a valid operation keyword. Valid values include: \
\ntables\nredactions\nfigures\nlayout\n"
)
return json.dumps(annotation, default=npconvert)
def make_annotations(pdf, annotation_function):
results = []
for i, page in enumerate(pdf):
boxes = annotation_function(page)
cells = [{"x": x, "y": y, "width": w, "height": h} for x, y, w, h in boxes]
results.append({"page": i, "pageWidth": page.shape[1], "pageHeight": page.shape[0], "cells": cells})
output_dict = {"pages": results}
return jsonify(json.dumps(output_dict, default=npconvert))
def get_size(data):
return round(getsizeof(data) / 1000000, 2)
def annotate(task):
def annotate(annotation_function):
def inner():
data = request.data
logger.info(f"Received data.")
logger.info(f"Processing data.")
pdf, angles = open_pdf(data)
# npages = len(pdf)
annotations = make_annotations(pdf, task)
return jsonify({"result": annotations, "deskew_angles": angles})
annotations = make_annotations(pdf, annotation_function)
# if CONFIG.deskew.function != "identity":
# annotations.update({"deskew_angles": angles})
return annotations
try:
return inner()