Coarse-to-Fine Lifted MAP Inference in Computer Vision

Authors: Haroun Habeeb, Ankit Anand, Mausam, Parag Singla

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the performance of C2F inference by developing lifted versions of two near state-of-the-art CV algorithms for stereo vision and interactive image segmentation.
Researcher Affiliation Academia Haroun Habeeb and Ankit Anand and Mausam and Parag Singla Indian Institute of Technology Delhi haroun7@gmail.com and {ankit.anand,mausam,parags}@cse.iitd.ac.in
Pseudocode Yes Algorithm 1 Coarse-to-Fine Lifted MAP Algorithm
Open Source Code Yes We release our implementation for wider use by the community.1https://github.com/dair-iitd/c2fi4cv/
Open Datasets Yes We use the benchmark Middlebury Stereo datasets of 2003, 2005 and 2006 [Scharstein and Szeliski, 2003; Hirschmuller and Scharstein, 2007]. For the 2003 dataset, quarter-size images are used and for others, third-size images are used. The label space is of size 85 (85 distinct disparity labels). The dataset used is provided with the implementation. It is a part of the MSRC V2 dataset.6
Dataset Splits No The paper mentions using benchmark datasets (Middlebury Stereo and MSRC V2) but does not provide specific details on how these datasets were split into training, validation, and test sets, or if standard splits were used and where they are defined.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions several software components and implementations used (e.g., Open GM2, TSGO implementation, Cooperative Graph Cuts), but it does not specify their version numbers, which are necessary for reproducible software dependency information.
Experiment Setup Yes For the 2003 dataset, quarter-size images are used and for others, third-size images are used. The label space is of size 85 (85 distinct disparity labels). C2F TSGO uses outputs from the sequence CP(1, 1), CP(2, 1), CP(3, 1) and then refines to the original MRF. Model refinement is triggered whenever energy hasn't decreased in the last four iterations of alpha expansion (this becomes the stopping criteria C in Algorithm 1).