Saved in:
Bibliographic Details
Main Authors: Ramachandran, Rahul, Kulkarni, Tejal, Sharma, Charchit, Vijaykeerthy, Deepak, Balasubramanian, Vineeth N
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2409.04041
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913492783595520
author Ramachandran, Rahul
Kulkarni, Tejal
Sharma, Charchit
Vijaykeerthy, Deepak
Balasubramanian, Vineeth N
author_facet Ramachandran, Rahul
Kulkarni, Tejal
Sharma, Charchit
Vijaykeerthy, Deepak
Balasubramanian, Vineeth N
contents Evaluating models and datasets in computer vision remains a challenging task, with most leaderboards relying solely on accuracy. While accuracy is a popular metric for model evaluation, it provides only a coarse assessment by considering a single model's score on all dataset items. This paper explores Item Response Theory (IRT), a framework that infers interpretable latent parameters for an ensemble of models and each dataset item, enabling richer evaluation and analysis beyond the single accuracy number. Leveraging IRT, we assess model calibration, select informative data subsets, and demonstrate the usefulness of its latent parameters for analyzing and comparing models and datasets in computer vision.
format Preprint
id arxiv_https___arxiv_org_abs_2409_04041
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On Evaluation of Vision Datasets and Models using Human Competency Frameworks
Ramachandran, Rahul
Kulkarni, Tejal
Sharma, Charchit
Vijaykeerthy, Deepak
Balasubramanian, Vineeth N
Computer Vision and Pattern Recognition
Evaluating models and datasets in computer vision remains a challenging task, with most leaderboards relying solely on accuracy. While accuracy is a popular metric for model evaluation, it provides only a coarse assessment by considering a single model's score on all dataset items. This paper explores Item Response Theory (IRT), a framework that infers interpretable latent parameters for an ensemble of models and each dataset item, enabling richer evaluation and analysis beyond the single accuracy number. Leveraging IRT, we assess model calibration, select informative data subsets, and demonstrate the usefulness of its latent parameters for analyzing and comparing models and datasets in computer vision.
title On Evaluation of Vision Datasets and Models using Human Competency Frameworks
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2409.04041