Torch Content Area

A PyTorch tool kit for estimating the circular content area in endoscopic footage. This implementation is released alongside our publication:

Rapid and robust endoscopic content area estimation: A lean GPU-based pipeline and curated benchmark dataset,
Charlie Budd, Luis C. Garcia-Peraza-Herrera, Martin Huber, Sebastien Ourselin, Tom Vercauteren.
[ arXiv ]

If you make use of this work, please cite the paper.

Installation

To install the latest version, simply run...

pip install torchcontentarea

Usage

from torchvision.io import read_image
from torchcontentarea import estimate_area, get_points, fit_area

# Image in NCHW format, byte/uint8 type is expected
image = read_image("my_image.png").unsqueeze(0)

# Either directly estimate area from image...
area = estimate_area(image, strip_count=16)

# ...or get the set of points and then fit the area.
points = get_points(image, strip_count=16)
area = fit_area(points, image.shape[2:4])

Performance

Performance is measured against the CholecECA subset of the Endoscopic Content Area (ECA) dataset.

Performance Results (handcrafted cpu)...

• Avg Time (Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz): 2.501ms
• Avg Error (Hausdorff Distance): 3.535
• Miss Rate (Error > 15): 2.1%
• Bad Miss Rate (Error > 25): 1.0%

Performance Results (learned cpu)...

• Avg Time (Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz): 4.662ms
• Avg Error (Hausdorff Distance): 4.388
• Miss Rate (Error > 15): 2.6%
• Bad Miss Rate (Error > 25): 1.4%

Performance Results (handcrafted cuda)...

• Avg Time (NVIDIA GeForce GTX 980 Ti): 0.171ms
• Avg Error (Hausdorff Distance): 4.289
• Miss Rate (Error > 15): 2.4%
• Bad Miss Rate (Error > 25): 1.3%

Performance Results (learned cuda)...

• Avg Time (NVIDIA GeForce GTX 980 Ti): 1.349ms
• Avg Error (Hausdorff Distance): 4.641
• Miss Rate (Error > 15): 2.6%
• Bad Miss Rate (Error > 25): 1.3%