Dataset Issues in Object Recognition

By Jean Ponce, Tamara L. Berg, Mark Everingham, D.A. Forsyth, Martial Hebert, Svetlana Lazebnik, Marcin Marszalek, Cordelia Schmid, Bryan C. Russell, Antonio Torralba and 3 more. 2006

262 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

In Evaluation & Benchmarks

Citations in the corpus

Do ImageNet Classifiers Generalize to ImageNet? (2019, neutral): “[45] Jean Ponce, Tamara L.”
Learning Visual Features from Large Weakly Supervised Data (2015, background): “…you re carrying a bag with some room it s nearly x x and weighs ounces new one is x x and weighs ounces i underexposed this one a bit did exposure bracketing script underexposure on that camera looks melty yummy gold kodak ﬁlm like not as impressive as embankment that s for sure ciﬁc task [37, 48].”
Recognizing indoor scenes (2009, background): “Note hat these averages have very few distinctive attributes in comparison with average images for the fifteen scene categories dataset and Caltech 101 [10].”
Dataset Distillation (2018, methodology): “Analyzing training datasets has, in the past, been mainly focused on the investigation of bias in datasets (Ponce et al., 2006; Torralba & Efros, 2011).”
Recognition in Terra Incognita (2018, background): “Human photographers are biased towards well-lit, wellfocused images where the subjects are centered in the frame [30, 31].”
Few-Shot Adversarial Domain Adaptation (2017, background): “Naively training a classifier on one dataset for testing on another is known to produce sub-optimal results, because an effect known as dataset bias [42, 57, 56] or covariate shift [51] occurs due to a difference in the distributions of the images between the datasets.”
Undoing the Damage of Dataset Bias (2012, background): “Unfortunately, experiments in [4, 5] strongly suggest the existence of various types of bias (e.”
Unified Deep Supervised Domain Adaptation and Generalization (2017, neutral): “Visual recognition algorithms are trained with data from a source domain, and when they are tested on a target domain with marginal distribution that differs from the one of the sources, we experience the visual domain adaptation (DA) problem (also known as dataset bias [48, 59, 58], or covariate sh”
An empirical study of context in object detection (2009, background): “Its role is to utilize the multitude of biases embedded in how we take pictures (framing [36], focus, subject matter), how we select datasets [30], how we gravitate towards visual clichés [34], and even how we name our children [8]!”
A High-Throughput Screening Approach to Discovering Good Forms of Biologically Inspired Visual Representation (2009, neutral): “Previous investigations [10–12,35,36] have raised”
Variable Selection (2019, background): “This integration appears to be supported by considerable physiological and psychophysical evidence [25, 26].”
Unbiased look at dataset bias (2011, unknown): “PASCAL Visual Object Classes (VOC) [6] was a reaction against the lax training and testing standards of previous datasets [14].”
Berkeley MHAD: A comprehensive Multimodal Human Action Database (2013, background): “Some of the issues pertaining to popular image datasets used by the computer vision community were previously discussed in [23].”
Spatial pyramid matching (2009, unknown): “One is the availability of large-scale training datasets gathered from the Web (Ponce et al., 2006).”
LabelMe: Online Image Annotation and Applications (2010, unknown): “D”
From ImageNet to Image Classification: Contextualizing Progress on Benchmarks (2020, background): “1We release our refined ImageNet annotations at https:// github.com/MadryLab/ImageNetMultiLabel. lenges (Ponce et al., 2006; Torralba & Efros, 2011; Everingham et al., 2010; Russakovsky et al., 2015).”
Web-Based Classifiers for Human Action Recognition (2012, background): “And as discussed in [46], the number of objects, as well as objects’ pose and scale vary quite a bit across retrieved images.”
How far can you get with a modern face recognition test set using only simple features? (2009, result): “We are clearly not the first to identity the importance of evaluationin driving progressin face and object recognition [ 12 ]; our results add to a long-standing process of evaluation and re-evaluation of how algorithms and systems are evaluated.”
Attribute-Based Detection of Unfamiliar Classes with Humans in the Loop (2013, background): “While this closedset approach has enabled scientific progress within the field, the datasets used may suffer from biases and restrictions that can be exploited by certain algorithms [30] [35].”
Protecting against evaluation overfitting in empirical reinforcement learning (2011, background): “Such concerns have led some to question whether performance improvements on benchmark environments reﬂect real progress in machine learning or simply an increase in prior knowledge about those environments [6], [7], [8].”