lzy/sci-gui-agent-benchmark
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'main' into xiaochuanli/addChromeExtensions

Browse Source
This commit is contained in:
Tianbao Xie
2024-03-08 20:45:49 +08:00
committed by GitHub
parent 62fd8feebb 4b841c199a
commit f01153cadd
109 changed files with 7196 additions and 172 deletions
Download Patch File Download Diff File Expand all files Collapse all files

156
desktop_env/evaluators/metrics/docs.py
View File

@@ -11,6 +11,7 @@ from docx.shared import RGBColor
from odf.opendocument import load
from odf.text import P
from odf.text import Span
from rapidfuzz import fuzz
from skimage.color import deltaE_ciede2000
from skimage.color import rgb2lab
@@ -57,6 +58,10 @@ def contains_page_break(docx_file):
def compare_docx_files(file1, file2, **options):
ignore_blanks = options.get('ignore_blanks', True)
ignore_case = options.get('ignore_case', False)
ignore_order = options.get('ignore_order', False)
content_only = options.get('content_only', False)
def get_paragraph_texts_odt(document):
paragraphs = document.getElementsByType(P)
paragraph_texts = []
@@ -79,20 +84,37 @@ def compare_docx_files(file1, file2, **options):
doc2 = Document(file2)
doc1_paragraphs = [p.text for p in doc1.paragraphs]
doc2_paragraphs = [p.text for p in doc2.paragraphs]
if ignore_order:
doc1_paragraphs = sorted(doc1_paragraphs)
doc2_paragraphs = sorted(doc2_paragraphs)
elif file1.endswith('.odt') and file2.endswith('.odt'):
doc1 = load(file1)
doc2 = load(file2)
doc1_paragraphs = get_paragraph_texts_odt(doc1)
doc2_paragraphs = get_paragraph_texts_odt(doc2)
if ignore_order:
doc1_paragraphs = sorted(doc1_paragraphs)
doc2_paragraphs = sorted(doc2_paragraphs)
else:
# Unsupported file types or mismatch
print("Unsupported file types or mismatch between file types.")
return 0
if content_only:
# Compare the content of the documents
text1 = re.sub(r'\s+', ' ', '\n'.join(doc1_paragraphs)).strip()
text2 = re.sub(r'\s+', ' ', '\n'.join(doc2_paragraphs)).strip()
if ignore_case:
text1, text2 = text1.lower(), text2.lower()
similarity = fuzz.ratio(text1, text2) / 100.0
return similarity
# Process and compare documents
if ignore_blanks:
text1 = re.sub(r'\s+', ' ', '\n'.join(doc1_paragraphs)).strip()
text2 = re.sub(r'\s+', ' ', '\n'.join(doc2_paragraphs)).strip()
if ignore_case:
text1, text2 = text1.lower(), text2.lower()
if text1 != text2:
return 0
else:
@@ -106,6 +128,8 @@ def compare_docx_files(file1, file2, **options):
print("in compare")
# Compare each paragraph
for p1, p2 in zip(doc1_paragraphs, doc2_paragraphs):
if ignore_case:
p1, p2 = p1.lower(), p2.lower()
if p1 != p2:
print(p1)
print(p2)
@@ -157,6 +181,44 @@ def compare_docx_tables(docx_file1, docx_file2):
return 1
from io import BytesIO
from PIL import Image
def compare_docx_images(docx_file1, docx_file2):
doc1 = Document(docx_file1)
doc2 = Document(docx_file2)
def extract_images(doc):
images = []
for rel in doc.part.rels.values():
if "image" in rel.reltype:
img_data = rel.target_part.blob
images.append(BytesIO(img_data))
return images
images1 = extract_images(doc1)
images2 = extract_images(doc2)
if len(images1) != len(images2):
return 0
for img1, img2 in zip(images1, images2):
if Image.open(img1).tobytes() != Image.open(img2).tobytes():
return 0
return 1
import pytesseract
def compare_image_text(image_path, rule):
img = Image.open(image_path)
img_text = pytesseract.image_to_string(img)
if rule['type'] == 'text':
return 1 if rule['text'] in img_text else 0
else:
raise ValueError("Unsupported rule type")
def compare_line_spacing(docx_file1, docx_file2):
if not compare_docx_files(docx_file1, docx_file2):
return 0
@@ -263,7 +325,7 @@ def check_tabstops(docx_file1, docx_file2, **kwargs) -> float:
section = doc2.sections[0]
paragraph_width = section.page_width - section.left_margin - section.right_margin
ignore_tabs = lambda x: x.alignment == WD_TAB_ALIGNMENT.CLEAR or (
x.alignment == WD_TAB_ALIGNMENT.LEFT and x.position == 0)
x.alignment == WD_TAB_ALIGNMENT.LEFT and x.position == 0)
minus = .0
for p1, p2 in zip(para1, para2):
# filter CLEAR tabstop and default left-0 tabstop
@@ -566,3 +628,95 @@ def compare_highlighted_text(file1, file2):
return 1
else:
return 0
def compare_references(file1, file2, **options):
reference_indicator = options.get('reference_indicator', 'References')
reference_base_result = options.get('reference_base_result', 0.5)
# Determine file types and load documents
if file1.endswith('.docx') and file2.endswith('.docx'):
doc1 = Document(file1)
doc2 = Document(file2)
doc1_paragraphs = [p.text for p in doc1.paragraphs]
doc2_paragraphs = [p.text for p in doc2.paragraphs]
else:
# Unsupported file types or mismatch
print("Unsupported file types or mismatch between file types.")
return 0
# Find the references section in the paragraphs, find the idx of the last reference_indicator in the paragraph list
ref1_idx = doc1_paragraphs.index(reference_indicator) if reference_indicator in doc1_paragraphs else -1
ref2_idx = doc2_paragraphs.index(reference_indicator) if reference_indicator in doc2_paragraphs else -1
if ref1_idx == -1 and ref2_idx == -1:
return 1
if ref1_idx == -1 or ref2_idx == -1:
return 0
# split the reference section into reference items, and remove the empty string items
ref1 = [p for p in doc1_paragraphs[ref1_idx + 1:] if p.strip()]
ref2 = [p for p in doc2_paragraphs[ref2_idx + 1:] if p.strip()]
# Compare the references
if len(ref1) != len(ref2):
return 0
total_similarity = 0
for r1, r2 in zip(ref1, ref2):
# fuzzy match the references
similarity = fuzz.ratio(r1, r2) / 100.0
total_similarity += similarity
result = total_similarity / len(ref1)
if result >= reference_base_result:
return (result - reference_base_result) / (1 - reference_base_result)
else:
return 0
def compare_answer(file1, file2, **options):
"""This is a specific function to compare the """
# Determine file types and load documents
if file1.endswith('.docx') and file2.endswith('.docx'):
doc1 = Document(file1)
doc2 = Document(file2)
doc1_paragraphs = [p.text for p in doc1.paragraphs]
doc2_paragraphs = [p.text for p in doc2.paragraphs]
else:
# Unsupported file types or mismatch
print("Unsupported file types or mismatch between file types.")
return 0
# Find the references section in the paragraphs, find the idx of the last reference_indicator in the paragraph list
ref1_idx = doc1_paragraphs.index(reference_indicator) if reference_indicator in doc1_paragraphs else -1
ref2_idx = doc2_paragraphs.index(reference_indicator) if reference_indicator in doc2_paragraphs else -1
if ref1_idx == -1 and ref2_idx == -1:
return 1
if ref1_idx == -1 or ref2_idx == -1:
return 0
# split the reference section into reference items, and remove the empty string items
ref1 = [p for p in doc1_paragraphs[ref1_idx + 1:] if p.strip()]
ref2 = [p for p in doc2_paragraphs[ref2_idx + 1:] if p.strip()]
# Compare the references
if len(ref1) != len(ref2):
return 0
total_similarity = 0
for r1, r2 in zip(ref1, ref2):
# fuzzy match the references
similarity = fuzz.ratio(r1, r2) / 100.0
total_similarity += similarity
result = total_similarity / len(ref1)
if result >= reference_base_result:
return (result - reference_base_result) / (1 - reference_base_result)
else:
return 0