A Deep Active Contour Model for Delineating Glacier Calving Fronts

TL;DR

This work completely rephrase the glacier calving front modeling task as a contour tracing problem and proposes a model for explicit contour detection that does not incorporate any dense predictions as intermediate steps, and demonstrates that explicit contours detection has benefits over pixel-wise methods when quantifying the models’ prediction uncertainties.

Abstract

Choosing how to encode a real-world problem as a machine learning task is an important design decision in machine learning. The task of glacier calving front modeling has often been approached as a semantic segmentation task. Recent studies have shown that combining segmentation with edge detection can improve the accuracy of calving front detectors. Building on this observation, we completely rephrase the task as a contour tracing problem and propose a model for explicit contour detection that does not incorporate any dense predictions as intermediate steps. The proposed approach, called ``Charting Outlines by Recurrent Adaptation'' (COBRA), combines Convolutional Neural Networks (CNNs) for feature extraction and active contour models for the delineation. By training and evaluating on several large-scale datasets of Greenland's outlet glaciers, we show that this approach indeed outperforms the aforementioned methods based on segmentation and edge-detection. Finally, we demonstrate that explicit contour detection has benefits over pixel-wise methods when quantifying the models' prediction uncertainties. The project page containing the code and animated model predictions can be found at \url{https://khdlr.github.io/COBRA/}.

Figures (5)

• Figure 1: High-level overview of our deep active contour model for delineating calving fronts. First, the backbone network takes the input image (a) and derives feature maps. Then, a sample is taken from these feature maps at the position of each vertex (b). These features are evaluated by the Snake Head (c) which predicts offsets for each vertex. Finally, the offsets are applied to update the contour (d). This process is repeated multiple times.
• Figure 2: Architecture overview of our model. Note that while this diagram shows only two iterations of the Snake Head, the number of iterations is actually an arbitrary hyperparameter, which we set to 4 for our experiments.
• Figure 3: Detailed view of the Snake Head. After the feature maps are sampled at the vertex positions, the vertex coordinates are concatenated to the vertex features. The 1D CNN then predicts offsets for each vertex. These offsets are added to the input coordinates to obtain the Snake Head's output.
• Figure 4: Visualization of prediction results (blue) for the different models and corresponding ground truth (red) on the test datasets. For the iterative, contour-based models intermediate results are displayed as dashed blue lines. Best viewed in color.
• Figure 5: Visualization of posterior samples obtained using Monte Carlo dropout (blue) from the different models overlaid on top of the ground truth (red) for scenes from the CALFIN test set.