Walk the Lines 2: Contour Tracking for Detailed Segmentation

TL;DR

Walk the Lines 2 (WtL2) tackles the need for highly detailed segmentation in niche domains by extending contour-tracking refinement to infrared ship imagery and a broad set of RGB objects. It achieves this by retraining the object contour detector for IR ships (RCN-IR) and adapting the tracking CNN and binarization to handle diverse categories, producing robust, closed, 1-pixel-wide contours suitable for precise foreground masks. The approach is evaluated against baselines and state-of-the-art contour methods on IR and RGB benchmarks, including the IR-specific ESIRRID/DIRSC data and a COCO-derived Detailed Object Contour (DOC) dataset, showing improved closed-contour formation and high IoU in favorable cases. These results suggest WtL2’s potential for high-detail segmentation in specialized applications, enabling precise samples and advancing niche areas such as IR maritime analysis and fine-grained object delineation.

Abstract

This paper presents Walk the Lines 2 (WtL2), a unique contour tracking algorithm specifically adapted for detailed segmentation of infrared (IR) ships and various objects in RGB.1 This extends the original Walk the Lines (WtL) [12], which focused solely on detailed ship segmentation in color. These innovative WtLs can replace the standard non-maximum suppression (NMS) by using contour tracking to refine the object contour until a 1-pixel-wide closed shape can be binarized, forming a segmentable area in foreground-background scenarios. WtL2 broadens the application range of WtL beyond its original scope, adapting to IR and expanding to diverse objects within the RGB context. To achieve IR segmentation, we adapt its input, the object contour detector, to IR ships. In addition, the algorithm is enhanced to process a wide range of RGB objects, outperforming the latest generation of contour-based methods when achieving a closed object contour, offering high peak Intersection over Union (IoU) with impressive details. This positions WtL2 as a compelling method for specialized applications that require detailed segmentation or high-quality samples, potentially accelerating progress in several niche areas of image segmentation.

Figures (6)

• Figure 1: Visualization of \ref{['orimoto']} the original, which, when processed with an object contour detector, produces a \ref{['soft']} soft contour with two post-processing options: \ref{['nmsdeer']} NMS, which inserts gaps in the contours (marked by red circles), or \ref{['wtlcontour']} WtL2-contour, which refines the contours and keeps them connected. Further processing results in the \ref{['wtl2bin']} 'perfect' binary WtL2. We visually compare contour-based segmentation, in particular \ref{['sotasegm']} E2EC with our \ref{['wtl2segm']} WtL2-seg.
• Figure 2: Diagram illustrating the contour tracking process of WtLs.
• Figure 3: Overview of WtL and WtL2 processes. Red outlines core algorithm. Modifications for WtL2 are outlined in color: adaptation for IR ship via object contour detector retraining in blue, and extension for various objects in green.
• Figure 4: Visualization of \ref{['image']} RGB segmentation method (no segment visible, like original image) \ref{['blueimage']} + strategies, \ref{['segmcrf']} + CRF, and \ref{['contour']} conversion to contour label; images from Kelm2025.
• Figure 5: Visualization of different IR ship segmentation results: (a) Original, (b) Ground Truth, (c) RefineNet-IR (baseline), (d) WtL2-seg-IR (ours); images from Kelm2025.