Finding Frequent Entities in Continuous Data

Authors: Ferran Alet, Rohan Chitnis, Leslie P. Kaelbling, Tomas Lozano-Perez

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We go on to describe the algorithm and its formal guarantees and describe experiments that find the main characters in video of a TV show and that address the household object-finding problem.
Researcher Affiliation Academia Ferran Alet, Rohan Chitnis, Leslie P. Kaelbling and Tom as Lozano-P erez MIT Computer Science and Artificial Intelligence Laboratory {alet, ronuchit, lpk, tlp}@mit.edu
Pseudocode Yes Algorithm 1: Hop And Count with multiple radii.
Open Source Code No The paper does not provide an explicit statement or link for the open-source code of the described methodology (HAC).
Open Datasets Yes We created a dataset of 8 humans moving objects around 20 different locations in a room.1 You can find it at http://lis.csail.mit.edu/alet/entities.html
Dataset Splits No The paper mentions using "70% of the data" for sampling in the House M.D. experiment, but does not specify clear train/validation/test splits for model training or evaluation. For the object localization experiment, no explicit splits are mentioned.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments.
Software Dependencies No The paper mentions using dlib, sklearn, Sharp Mask, and Inception-V3, but does not specify version numbers for these software dependencies.
Experiment Setup Yes We run HAC with r = 0.5 and apply duplicate reduction with rd = 0.65. These parameters were not fine-tuned; they were picked based on comments from the paper that created the CNN and on figure 6. We fix ϵ = δ = 0.5 for all experiments; these large values are sufficient because HAC works better in high dimensions than guaranteed by theorem 2.1. Run HAC with τl = (all points have the same weight regardless of their time), τs = 10 seconds, f = 2.5% and a distance function and threshold which link two detections that happen roughly within 30 centimeters and have features that are close in embedding space.