Planar Ultrametrics for Image Segmentation

Authors: Julian E. Yarkony, Charless Fowlkes

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our algorithm to the task of natural image segmentation and demonstrate that our algorithm converges rapidly and produces optimal or near-optimal solutions in practice. We applied our algorithm to segmenting images from the Berkeley Segmentation Data set (BSDS) [16].
Researcher Affiliation Collaboration Julian Yarkony Experian Data Lab San Diego, CA 92130 julian.yarkony@experian.com Charless C. Fowlkes Department of Computer Science University of California Irvine fowlkes@ics.uci.edu
Pseudocode Yes Algorithm 1 Dual Closest Ultrametric via Cutting Planes
Open Source Code No The paper mentions using existing libraries (Blossom V, IBM ILOG CPLEX) but does not state that their own code for the described methodology is open-source or provide a link.
Open Datasets Yes We applied our algorithm to segmenting images from the Berkeley Segmentation Data set (BSDS) [16].
Dataset Splits No The paper references the BSDS dataset but does not explicitly provide details about training, validation, or test dataset splits (percentages, counts, or specific predefined split usage).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions "MATLAB", "Blossom V implementation of minimum-weight perfect matching [15]" and "IBM ILOG CPLEX LP solver" but does not provide specific version numbers for these software components.
Experiment Setup Yes We truncate extreme values to enforce that g Pbe [ϵ, 1 ϵ] with ϵ = 0.001 and set θe = log g P be 1 g P be. In our experiments we use a fixed set of eleven distance threshold levels {δl} chosen to uniformly span the useful range of threshold values [9.6, 12.6]. We performed dual cutting plane iterations until convergence or 2000 seconds had passed. ... We terminate when the total residual is greater than 2 10 4. In our experiments, this penalty is scaled by a parameter ϵ = 10 4. In our implementation we test a few discrete thresholds t {0, 0.2, 0.4, 0.6, 0.8} and take that threshold that yields X with the lowest cost.