toon3d / app.py
ethanweber's picture
update
02b79da
raw
history blame
11.5 kB
import spaces
import gradio as gr
import os
import shutil
from pathlib import Path
from toon3d.scripts.viser_vis import main as viser_vis_main
import viser
import time
import threading
viewer_thread_instance = None
stop_event = threading.Event()
shared_url = None
_HEADER_ = '''
<h2>Toon3D: Seeing Cartoons from a New Perspective</h2>
Toon3D lifts hand-drawn images to 3D with a piecewise-rigid deformation optimization at hand-labeled keypoints and using monocular depth as a prior. The project page is at <a href='https://toon3d.studio/' target='_blank'>https://toon3d.studio/</a> and the Toon3D Labeler is at <a href='https://labeler.toon3d.studio/' target='_blank'>https://labeler.toon3d.studio/</a>. Follow the steps below to run Toon3D!
<div style="margin-top: 20px; font-size: 16px; line-height: 1.6;">
<div style="display: flex; justify-content: space-between;">
<div style="width: 49%;">
<ol>
<li><strong>Prepare and Process Data</strong>
<ul>
<li>Upload images and click on "Process Data" to generate processed data.</li>
<li>Download the processed data.</li>
</ul>
</li>
<li><strong>Label Data</strong>
<ul>
<li>Upload the processed data and label points using the labeler ("Upload ZIP").</li>
<li>Click export and upload the points.json to the "Labeled Points" section.</li>
</ul>
</li>
</ol>
</div>
<div style="width: 49%;">
<ol start="3">
<li><strong>Generate 3D Output</strong>
<ul>
<li>Click on "Run Toon3D" to run the structure from motion pipeline.</li>
<li>Download the output and inspect locally (point cloud, mesh, Nerfstudio dataset).</li>
</ul>
</li>
<li><strong>View in Web!</strong>
<ul>
<li>Click on "Open Viewer" to view the output in an interactive viewer powered by <a href="https://viser.studio/">Viser</a>.</li>
</ul>
<ul>
<li>Reach out if you have any questions!</li>
</ul>
</li>
</ol>
</div>
</div>
</div>
'''
def check_input_images(input_images):
if input_images is None:
raise gr.Error("No images uploaded!")
@spaces.GPU(duration=120)
def process_images(input_images, compute_segment_anything):
images_path = "/tmp/gradio/images"
processed_path = "/tmp/gradio/processed"
# remove the images_path folder
os.system(f"rm -rf {images_path}")
os.system(f"rm -rf {processed_path}")
# copy the uploaded images to the images_path folder
os.system(f"mkdir -p {images_path}")
os.system(f"mkdir -p {processed_path}")
for fileobj in input_images:
shutil.copyfile(fileobj.name, images_path + "/" + os.path.basename(fileobj.name))
# download SAM checkpoint
download_cmd = "tnd-download-data sam --save-dir /tmp/gradio"
os.system(download_cmd)
if compute_segment_anything:
sam_cmd = " --compute-segment-anything"
else:
sam_cmd = " --no-compute-segment-anything"
# process the data
process_data_cmd = f"tnd-process-data initialize --dataset toon3d-dataset --input_path {images_path} --data_prefix {processed_path} --sam_checkpoint_prefix /tmp/gradio/sam-checkpoints"
process_data_cmd += sam_cmd
print(process_data_cmd)
os.system(process_data_cmd)
zip_folder = "/tmp/gradio/processed/toon3d-dataset"
shutil.make_archive(zip_folder, 'zip', zip_folder)
return zip_folder + ".zip"
def toggle_labeler_visibility(visible):
if visible:
return '<iframe src="https://labeler.toon3d.studio/" style="display: block; margin: auto; width: 100%; height: 100vh;" frameborder="0"></iframe>'
else:
return ""
def check_input_toon3d(processed_data_zip, labeled_data):
if processed_data_zip is None:
raise gr.Error("No images uploaded!")
@spaces.GPU(duration=120)
def run_toon3d(processed_data_zip, labeled_data):
data_prefix = "/tmp/gradio/inputs"
processed_path = f"{data_prefix}/toon3d-dataset"
output_prefix = "/tmp/gradio/outputs"
nerfstudio_folder = "/tmp/gradio/nerfstudio"
os.system(f"rm -rf {processed_path}")
os.system(f"rm -rf {output_prefix}")
os.system(f"rm -rf {nerfstudio_folder}")
shutil.unpack_archive(processed_data_zip.name, processed_path)
shutil.copyfile(labeled_data.name, f"{processed_path}/points.json")
# run toon3d
toon3d_cmd = f"tnd-run --dataset toon3d-dataset --data_prefix {data_prefix} --output_prefix {output_prefix} --nerfstudio_folder {nerfstudio_folder} --no-view-point-cloud"
os.system(toon3d_cmd)
# get the last timestamped folder in output_prefix
# output_folder = sorted([f.path for f in os.scandir(output_prefix) if f.is_dir()])[-1]
output_dirs = Path(output_prefix) / "toon3d-dataset" / "run"
output_dir = Path(output_dirs / sorted(os.listdir(output_dirs))[-1])
zip_folder = str(output_dir)
shutil.make_archive(zip_folder, 'zip', zip_folder)
return zip_folder + ".zip"
# def open_viewer_fn(processed_data_zip, labeled_data, toon3d_output_zip):
# print(processed_data_zip)
# print(labeled_data)
# print(toon3d_output_zip)
# data_prefix = Path("/tmp/gradio/inputs")
# processed_path = f"{data_prefix}/toon3d-dataset"
# # extract the zip file
# viewer_folder = "/tmp/gradio/viewer/toon3d-dataset/run/temp"
# os.system(f"rm -rf {viewer_folder}")
# shutil.unpack_archive(toon3d_output_zip.name, viewer_folder)
# shutil.unpack_archive(processed_data_zip.name, processed_path)
# shutil.copyfile(labeled_data.name, f"{processed_path}/points.json")
# viser_server = viser.ViserServer()
# url = viser_server.request_share_url()
# print(url)
# # this is an infinite while loop so needs to be run in a separate thread
# # TODO:
# viser_vis_main(
# data_prefix=data_prefix,
# dataset="toon3d-dataset",
# output_prefix=Path("/tmp/gradio/viewer"),
# output_method=Path("run"),
# server=viser_server,
# visible=True,
# )
def viewer_thread(processed_data_zip, labeled_data, toon3d_output_zip):
global shared_url
data_prefix = Path("/tmp/gradio/inputs")
processed_path = f"{data_prefix}/toon3d-dataset"
viewer_folder = "/tmp/gradio/viewer/toon3d-dataset/run/temp"
os.system(f"rm -rf {viewer_folder}")
shutil.unpack_archive(toon3d_output_zip.name, viewer_folder)
shutil.unpack_archive(processed_data_zip.name, processed_path)
shutil.copyfile(labeled_data.name, f"{processed_path}/points.json")
viser_server = viser.ViserServer()
url = viser_server.request_share_url()
shared_url = url # Save the URL to the global variable
print(url)
viser_vis_main(
data_prefix=data_prefix,
dataset="toon3d-dataset",
output_prefix=Path("/tmp/gradio/viewer"),
output_method=Path("run"),
server=viser_server,
visible=True,
return_early=True
)
while not stop_event.is_set():
time.sleep(1)
viser_server.stop() # Ensure the server is stopped when the loop exits
def kill_viewer():
global viewer_thread_instance, stop_event
if viewer_thread_instance and viewer_thread_instance.is_alive():
stop_event.set() # Signal the thread to stop
viewer_thread_instance.join() # Wait for the thread to actually stop
viewer_thread_instance = None
print("Viewer has been stopped.")
else:
print("No viewer is running.")
def get_html_for_shared_url(url):
return f'<h1>Open <a href="{url}" target="_blank">{url}</a>!</h1>'
def check_input_open_viewer(processed_data_zip, labeled_data, toon3d_output_zip):
if processed_data_zip is None:
raise gr.Error("No processed data uploaded!")
if labeled_data is None:
raise gr.Error("No labeled points uploaded!")
if toon3d_output_zip is None:
raise gr.Error("No Toon3D output uploaded!")
def start_viewer(processed_data_zip, labeled_data, toon3d_output_zip):
kill_viewer() # Kill the existing viewer if it's running
global viewer_thread_instance, stop_event, shared_url
stop_event.clear() # Reset the stop event
shared_url = None # Reset the URL before starting
if viewer_thread_instance is None or not viewer_thread_instance.is_alive():
viewer_thread_instance = threading.Thread(target=viewer_thread, args=(processed_data_zip, labeled_data, toon3d_output_zip))
viewer_thread_instance.start()
while not shared_url:
# Wait for the URL to be set by the thread
time.sleep(0.1)
return get_html_for_shared_url(shared_url) # Return the URL after the thread has set it
else:
print("Viewer is already running.")
return get_html_for_shared_url(shared_url) # Return the current URL if the viewer is already running
with gr.Blocks(title="Toon3D") as demo:
gr.Markdown(_HEADER_)
with gr.Row(variant="panel"):
input_images = gr.File(label="Upload Images", file_count="multiple", file_types=[".jpg", "jpeg", "png"])
with gr.Column():
compute_segment_anything = gr.Checkbox(label="Compute Segment Anything? (slow)", value=False)
process_data_button = gr.Button("Process Data", elem_id="process_data_button", variant="primary")
processed_data_zip = gr.File(label="Processed Data", file_count="single", file_types=[".zip"], interactive=True)
with gr.Row(variant="panel"):
labeler_visible = gr.Checkbox(label="Show Labeler", value=False)
with gr.Row(variant="panel"):
labeler_frame = gr.HTML()
labeler_visible.change(toggle_labeler_visibility, inputs=[labeler_visible], outputs=[labeler_frame])
with gr.Row(variant="panel"):
labeled_data = gr.File(label="Labeled Points", file_count="single", file_types=[".json"])
run_toon3d_button = gr.Button("Run Toon3D", elem_id="run_toon3d_button", variant="primary")
toon3d_output_zip = gr.File(label="Toon3D Output", file_count="single", file_types=[".zip"], interactive=True)
with gr.Row(variant="panel"):
open_viewer_button = gr.Button("Open Viewer", elem_id="open_viser_button", variant="primary")
with gr.Row(variant="panel"):
viser_link = gr.HTML()
process_data_button.click(fn=check_input_images, inputs=[input_images]).success(
fn=process_images,
inputs=[input_images, compute_segment_anything],
outputs=[processed_data_zip],
)
run_toon3d_button.click(fn=check_input_toon3d, inputs=[processed_data_zip, labeled_data]).success(
fn=run_toon3d,
inputs=[processed_data_zip, labeled_data],
outputs=[toon3d_output_zip],
)
open_viewer_button.click(fn=check_input_open_viewer, inputs=[processed_data_zip, labeled_data, toon3d_output_zip]).success(
fn=start_viewer,
inputs=[processed_data_zip, labeled_data, toon3d_output_zip],
outputs=[viser_link],
)
if __name__ == "__main__":
demo.queue(max_size=10)
demo.launch()