Submodular meets Structured: Finding Diverse Subsets in Exponentially-Large Structured Item Sets

Authors: Adarsh Prasad, Stefanie Jegelka, Dhruv Batra

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical contribution is an efficient algorithm for producing a set of image segmentations with significantly higher oracle accuracy than previous works. We apply our greedy maximization algorithms to two image segmentation problems: (1) interactive binary segmentation (object cutout) (Section 4.1); (2) category-level object segmentation on the PASCAL VOC 2012 dataset [11] (Section 4.2). We compare all methods by their respective oracle accuracies, i.e. the accuracy of the most accurate segmentation in the set of M diverse segmentations returned by that method.
Researcher Affiliation Academia Adarsh Prasad UT Austin adarsh@cs.utexas.edu Stefanie Jegelka UC Berkeley stefje@eecs.berkeley.edu Dhruv Batra Virginia Tech dbatra@vt.edu
Pseudocode No The paper describes algorithmic steps verbally, but does not include formal pseudocode blocks or algorithms labeled as such.
Open Source Code No The paper does not provide an explicit statement or link to its own open-source code for the described methodology. It mentions using 'Cyborg implementation' and referring to other works' implementations for inference, but not its own overall code.
Open Datasets Yes We apply our greedy maximization algorithms to two image segmentation problems: (1) interactive binary segmentation (object cutout) (Section 4.1); (2) category-level object segmentation on the PASCAL VOC 2012 dataset [11] (Section 4.2).
Dataset Splits Yes We evaluate all methods on the PASCAL VOC 2012 data [11], consisting of train, val and test partitions with about 1450 images each.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments (e.g., CPU, GPU models, memory).
Software Dependencies No The paper mentions using specific implementations like the 'Cyborg implementation' [30], 'implementations of Delong et al. [9]', and 'Boykov et al. [3]', but does not provide version numbers for any of these software components.
Experiment Setup Yes Fifty of the images were used for tuning the diversity parameters λ, γ, and the other 50 for reporting oracle accuracies. The 2-label contrast-sensitive Potts model results in a supermodular relevance function r(y), which can be efficiently maximized via graph cuts [20]. The Hamming ball diversity dlb(y|S) is a collection of cardinality factors, which we optimize with the Cyborg implementation [30]. We construct a multi-label pairwise CRF on superpixels. Our node potentials are outputs of category-specific regressors trained by [6], and our edge potentials are multi-label Potts. Inference in the presence of diversity terms is performed with the implementations of Delong et al. [9] for label costs, Tarlow et al. [30] for Hamming ball diversity, and Boykov et al. [3] for label transitions. Diversity parameters (γ, λ) are chosen by performing cross-val on val.