vision/m4/sourcing/data_collection/visualization/global_visualization.py [72:267]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
)
self.extractor = TextMediaPairsExtractor(
dom_tree_simplificator=self.dom_tree_simplificator,
pre_extraction_simplificator=self.pre_extraction_simplificator_merge_texts,
also_extract_images_not_in_simplified_dom_tree=True,
extract_clip_scores=True,
)
def visualization(self):
st.title(
"Visualization of DOM tree simplification strategies, "
"web document rendering, and text-image pair extractions"
)
self.choose_mode()
self.choose_example()
self.simplification_mode()
self.extraction_mode()
def choose_mode(self):
st.header("Mode")
self.mode = st.selectbox(
label="Choose a mode",
options=["Simplification", "Extraction"],
index=1,
)
def choose_example(self):
st.header("Document")
if st.button("Select a random document"):
dct_idx = random.randint(a=0, b=self.num_docs - 1)
else:
dct_idx = 0
idx = st.number_input(
f"Select a document among the first {self.num_docs} ones",
min_value=0,
max_value=self.num_docs - 1,
value=dct_idx,
step=1,
help=f"Index between 0 and {self.num_docs-1}",
)
self.current_example = self.examples[idx]
def get_dom_viz_html(self, html):
def get_body_html_string(html):
tree = make_selectolax_tree(html)
tree.strip_tags(["script"])
return tree.body.html
body_html = get_body_html_string(html)
rendered_dom = self.dom_viz_template.render(body_html=body_html)
return rendered_dom
def simplification_mode(self):
if self.mode == "Simplification":
current_html = self.current_example["html"]
current_url = self.current_example["url"]
simplified_current_html = self.dom_tree_simplificator(current_html, type_return="str")
def display_rendered_webpages():
st.header("Rendered webpage")
st.markdown(f"Webpage url: {current_url}")
col1, col2 = st.columns(2)
with col1:
st.subheader("Raw html rendering")
st.components.v1.html(current_html, height=450, scrolling=True)
with col2:
st.subheader("Simplified html rendering")
st.components.v1.html(simplified_current_html, height=450, scrolling=True)
def display_dom_trees():
st.header("DOM trees")
col1, col2 = st.columns(2)
with col1:
st.subheader("Raw DOM tree")
rendered_dom = self.get_dom_viz_html(current_html)
st.components.v1.html(rendered_dom, height=600, scrolling=True)
with col2:
st.subheader("Simplified DOM tree")
simplified_rendered_dom = self.get_dom_viz_html(simplified_current_html)
st.components.v1.html(simplified_rendered_dom, height=600, scrolling=True)
def display_html_codes():
st.header("HTML codes")
col1, col2 = st.columns(2)
with col1:
st.subheader("Raw HTML code")
st.components.v1.html("<![CDATA[" + current_html + "]]>", height=450, scrolling=True)
with col2:
st.subheader("Simplified HTML code")
st.components.v1.html("<![CDATA[" + simplified_current_html + "]]>", height=450, scrolling=True)
display_rendered_webpages()
display_dom_trees()
display_html_codes()
def extraction_mode(self):
if self.mode == "Extraction":
current_html = self.current_example["html"]
current_url = self.current_example["url"]
simplified_current_html_tree = self.dom_tree_simplificator(current_html, type_return="selectolax_tree")
simplified_current_html = simplified_current_html_tree.html
current_list_nodes_not_merge_texts = self.pre_extraction_simplificator_not_merge_texts(
simplified_current_html_tree, page_url=current_url
)
current_list_nodes_merge_texts = self.pre_extraction_simplificator_merge_texts(
simplified_current_html_tree, page_url=current_url
)
extracted_images = self.extractor(html_str=current_html, page_url=current_url)
# For simplicity, only doing this replacement on the extracted images.
# Doing that before the extraction (i.e. in the DOM simplification) would be possible be would require
# more significant changes
replacement_dict = {
elem["unformatted_src"]: elem["src"]
for elem in extracted_images
if elem["src"] != elem["unformatted_src"]
}
def replace_relative_paths(html_string):
if replacement_dict:
for k, v in replacement_dict.items():
html_string = html_string.replace(k, v)
return html_string
def display_rendered_webpages():
st.header("Rendered webpage")
st.markdown(f"Webpage url: {current_url}")
display_raw_html_rendering = st.checkbox("Raw html rendering", value=True)
display_simplified_html_rendering = st.checkbox("Simplified html rendering", value=True)
col1, col2 = st.columns(2)
with col1:
display_pre_extraction_visualization = st.checkbox(
"Web document rendering (pre-extraction visualization)", value=True
)
with col2:
if display_pre_extraction_visualization:
merge_text_nodes = st.checkbox("Merge text nodes", value=True)
list_display_pages = [
[display_raw_html_rendering, "raw_html_rendering"],
[display_simplified_html_rendering, "simplified_html_rendering"],
[display_pre_extraction_visualization, "pre_extraction_visualization"],
]
list_display_pages = [
page_to_display
for should_display_page, page_to_display in list_display_pages
if should_display_page
]
def display_specific_rendered_webpage(page_to_display, col):
with col:
if page_to_display == "raw_html_rendering":
st.subheader("Raw html rendering")
st.components.v1.html(replace_relative_paths(current_html), height=800, scrolling=True)
elif page_to_display == "simplified_html_rendering":
st.subheader("Simplified html rendering")
st.components.v1.html(
replace_relative_paths(simplified_current_html), height=800, scrolling=True
)
elif page_to_display == "pre_extraction_visualization":
st.subheader("Web document rendering (pre-extraction visualization)")
def list_nodes_to_visu():
if not merge_text_nodes:
list_nodes = current_list_nodes_not_merge_texts
reduce_levels = {
v: i + 1
for i, v in enumerate(sorted(list(set([node.level for node in list_nodes]))))
}
last_level = None
markdown = ""
for node in list_nodes:
if node.tag in ["-text", "img"]:
current_level = reduce_levels[node.level]
if last_level != current_level:
markdown += (
"#" * min(current_level, 6) + f" Level: {current_level}\n\n"
)
last_level = current_level
path_in_tree_str = [tag for tag, _ in node.path_in_tree]
markdown += f"**{'/'.join(path_in_tree_str)}**\n\n"
if node.tag == "-text":
markdown += f"{node.text}\n\n"
elif node.tag == "img":
markdown += f"\n\n"
st.markdown(markdown)
else:
list_nodes = current_list_nodes_merge_texts
for node in list_nodes:
if node.tag == "-text":
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
vision/m4/sourcing/data_collection/visualization/wikipedia/global_visualization.py [71:266]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
)
self.extractor = TextMediaPairsExtractor(
dom_tree_simplificator=self.dom_tree_simplificator,
pre_extraction_simplificator=self.pre_extraction_simplificator_merge_texts,
also_extract_images_not_in_simplified_dom_tree=True,
extract_clip_scores=True,
)
def visualization(self):
st.title(
"Visualization of DOM tree simplification strategies, "
"web document rendering, and text-image pair extractions"
)
self.choose_mode()
self.choose_example()
self.simplification_mode()
self.extraction_mode()
def choose_mode(self):
st.header("Mode")
self.mode = st.selectbox(
label="Choose a mode",
options=["Simplification", "Extraction"],
index=1,
)
def choose_example(self):
st.header("Document")
if st.button("Select a random document"):
dct_idx = random.randint(a=0, b=self.num_docs - 1)
else:
dct_idx = 0
idx = st.number_input(
f"Select a document among the first {self.num_docs} ones",
min_value=0,
max_value=self.num_docs - 1,
value=dct_idx,
step=1,
help=f"Index between 0 and {self.num_docs-1}",
)
self.current_example = self.examples[idx]
def get_dom_viz_html(self, html):
def get_body_html_string(html):
tree = make_selectolax_tree(html)
tree.strip_tags(["script"])
return tree.body.html
body_html = get_body_html_string(html)
rendered_dom = self.dom_viz_template.render(body_html=body_html)
return rendered_dom
def simplification_mode(self):
if self.mode == "Simplification":
current_html = self.current_example["html"]
current_url = self.current_example["url"]
simplified_current_html = self.dom_tree_simplificator(current_html, type_return="str")
def display_rendered_webpages():
st.header("Rendered webpage")
st.markdown(f"Webpage url: {current_url}")
col1, col2 = st.columns(2)
with col1:
st.subheader("Raw html rendering")
st.components.v1.html(current_html, height=450, scrolling=True)
with col2:
st.subheader("Simplified html rendering")
st.components.v1.html(simplified_current_html, height=450, scrolling=True)
def display_dom_trees():
st.header("DOM trees")
col1, col2 = st.columns(2)
with col1:
st.subheader("Raw DOM tree")
rendered_dom = self.get_dom_viz_html(current_html)
st.components.v1.html(rendered_dom, height=600, scrolling=True)
with col2:
st.subheader("Simplified DOM tree")
simplified_rendered_dom = self.get_dom_viz_html(simplified_current_html)
st.components.v1.html(simplified_rendered_dom, height=600, scrolling=True)
def display_html_codes():
st.header("HTML codes")
col1, col2 = st.columns(2)
with col1:
st.subheader("Raw HTML code")
st.components.v1.html("<![CDATA[" + current_html + "]]>", height=450, scrolling=True)
with col2:
st.subheader("Simplified HTML code")
st.components.v1.html("<![CDATA[" + simplified_current_html + "]]>", height=450, scrolling=True)
display_rendered_webpages()
display_dom_trees()
display_html_codes()
def extraction_mode(self):
if self.mode == "Extraction":
current_html = self.current_example["html"]
current_url = self.current_example["url"]
simplified_current_html_tree = self.dom_tree_simplificator(current_html, type_return="selectolax_tree")
simplified_current_html = simplified_current_html_tree.html
current_list_nodes_not_merge_texts = self.pre_extraction_simplificator_not_merge_texts(
simplified_current_html_tree, page_url=current_url
)
current_list_nodes_merge_texts = self.pre_extraction_simplificator_merge_texts(
simplified_current_html_tree, page_url=current_url
)
extracted_images = self.extractor(html_str=current_html, page_url=current_url)
# For simplicity, only doing this replacement on the extracted images.
# Doing that before the extraction (i.e. in the DOM simplification) would be possible be would require
# more significant changes
replacement_dict = {
elem["unformatted_src"]: elem["src"]
for elem in extracted_images
if elem["src"] != elem["unformatted_src"]
}
def replace_relative_paths(html_string):
if replacement_dict:
for k, v in replacement_dict.items():
html_string = html_string.replace(k, v)
return html_string
def display_rendered_webpages():
st.header("Rendered webpage")
st.markdown(f"Webpage url: {current_url}")
display_raw_html_rendering = st.checkbox("Raw html rendering", value=True)
display_simplified_html_rendering = st.checkbox("Simplified html rendering", value=True)
col1, col2 = st.columns(2)
with col1:
display_pre_extraction_visualization = st.checkbox(
"Web document rendering (pre-extraction visualization)", value=True
)
with col2:
if display_pre_extraction_visualization:
merge_text_nodes = st.checkbox("Merge text nodes", value=True)
list_display_pages = [
[display_raw_html_rendering, "raw_html_rendering"],
[display_simplified_html_rendering, "simplified_html_rendering"],
[display_pre_extraction_visualization, "pre_extraction_visualization"],
]
list_display_pages = [
page_to_display
for should_display_page, page_to_display in list_display_pages
if should_display_page
]
def display_specific_rendered_webpage(page_to_display, col):
with col:
if page_to_display == "raw_html_rendering":
st.subheader("Raw html rendering")
st.components.v1.html(replace_relative_paths(current_html), height=800, scrolling=True)
elif page_to_display == "simplified_html_rendering":
st.subheader("Simplified html rendering")
st.components.v1.html(
replace_relative_paths(simplified_current_html), height=800, scrolling=True
)
elif page_to_display == "pre_extraction_visualization":
st.subheader("Web document rendering (pre-extraction visualization)")
def list_nodes_to_visu():
if not merge_text_nodes:
list_nodes = current_list_nodes_not_merge_texts
reduce_levels = {
v: i + 1
for i, v in enumerate(sorted(list(set([node.level for node in list_nodes]))))
}
last_level = None
markdown = ""
for node in list_nodes:
if node.tag in ["-text", "img"]:
current_level = reduce_levels[node.level]
if last_level != current_level:
markdown += (
"#" * min(current_level, 6) + f" Level: {current_level}\n\n"
)
last_level = current_level
path_in_tree_str = [tag for tag, _ in node.path_in_tree]
markdown += f"**{'/'.join(path_in_tree_str)}**\n\n"
if node.tag == "-text":
markdown += f"{node.text}\n\n"
elif node.tag == "img":
markdown += f"\n\n"
st.markdown(markdown)
else:
list_nodes = current_list_nodes_merge_texts
for node in list_nodes:
if node.tag == "-text":
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -