Recognition in Terra Incognita

Beery, Sara; Horn, Grant Van; Perona, Pietro

doi:10.1007/978-3-030-01270-0_28

Cited by 400 publications

(410 citation statements)

References 55 publications

Supporting

Mentioning

402

Contrasting

Order By: Relevance

“…However, this is but a first step and there are many questions still to be answered, such as the identification and mitigation of biases in social media imagery, and the propagation of styles across space. The problem of analyzing trends is also relevant in other visual domains, such as understanding which animals are getting rarer over time in camera trap images [3] or how land-use patterns are changing in satellite imagery [17]. We therefore believe that this is an important problem deserving of future research.…”

Section: Discussionmentioning

confidence: 99%

GeoStyle: Discovering Fashion Trends and Events

Mall

Matzen

Hariharan

et al. 2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

Understanding fashion styles and trends is of great potential interest to retailers and consumers alike. The photos people upload to social media are a historical and public data source of how people dress across the world and at different times. While we now have tools to automatically recognize the clothing and style attributes of what people are wearing in these photographs, we lack the ability to analyze spatial and temporal trends in these attributes or make predictions about the future. In this paper we address this need by providing an automatic framework that analyzes large corpora of street imagery to (a) discover and forecast longterm trends of various fashion attributes as well as automatically discovered styles, and (b) identify spatio-temporally localized events that affect what people wear. We show that our framework makes long term trend forecasts that are > 20% more accurate than prior art, and identifies hundreds of socially meaningful events that impact fashion across the globe. The supplementary material can be found at

show abstract

Section: Discussionmentioning

confidence: 99%

GeoStyle: Discovering Fashion Trends and Events

Mall

Matzen

Hariharan

et al. 2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

show abstract

“…In this regard, there are at least two sources of concern. First, they can have limited generalization beyond their training domain [32,2]. This is classically known as dataset bias.…”

Section: Introductionmentioning

confidence: 99%

REPAIR: Removing Representation Bias by Dataset Resampling

Vasconcelos

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

195

126

View full text Add to dashboard Cite

Modern machine learning datasets can have biases for certain representations that are leveraged by algorithms to achieve high performance without learning to solve the underlying task. This problem is referred to as "representation bias". The question of how to reduce the representation biases of a dataset is investigated and a new dataset REPresentAtion bIas Removal (REPAIR) procedure is proposed. This formulates bias minimization as an optimization problem, seeking a weight distribution that penalizes examples easy for a classifier built on a given feature representation. Bias reduction is then equated to maximizing the ratio between the classification loss on the reweighted dataset and the uncertainty of the ground-truth class labels. This is a minimax problem that REPAIR solves by alternatingly updating classifier parameters and dataset resampling weights, using stochastic gradient descent. An experimental set-up is also introduced to measure the bias of any dataset for a given representation, and the impact of this bias on the performance of recognition models. Experiments with synthetic and action recognition data show that dataset REPAIR can significantly reduce representation bias, and lead to improved generalization of models trained on REPAIRed datasets. The tools used for characterizing representation bias, and the proposed dataset RE-PAIR algorithm, are available at https://github.com/ JerryYLi/Dataset-REPAIR/.

show abstract

“…An instance of the effective integration of citizen scientist in deep learning is Wildlife@Home, where citizen science classifications were used to train neural networks that help to analyse bird populations [84]. The growing set of citizen science classified datasets, in addition to several research teams' datasets [85,86], are also being used to create software packages to aid ecological projects. Examples of this software are the 'R' package 'Machine Learning for Wildlife Image Classification' [4] and the pretrained networks in 'ClassifyMe' [87].…”

Section: Integrating Ai Into Camera Trap and Citizen Science Work Flowsmentioning

confidence: 99%

“…Nevertheless, for studies with small training data sets relative to the problem, accuracy may drop significantly [81]. Even where accuracy is high there is still an issue of generalisability, not captured in this broad performance metric, where a model performs worse on data on which it has not been specifically trained [85]. For example, performance might be substantially lower when a network trained to recognise a particular species is given the task of doing so against a novel habitat background.…”

Section: Integrating Ai Into Camera Trap and Citizen Science Work Flowsmentioning

confidence: 99%

Innovations in Camera Trapping Technology and Approaches: The Integration of Citizen Science and Artificial Intelligence

Green

Rees

Stephens

et al. 2020

Animals

View full text Add to dashboard Cite

Simple Summary: Camera traps, also known as "game cameras" or "trail cameras", have increasingly been used in wildlife research over the last 20 years. Although early units were bulky and the set-up was complicated, modern camera traps are compact, integrated units able to collect vast digital datasets. Some of the challenges now facing researchers include the time required to view, classify, and sort all of the footage collected, as well as the logistics of establishing and maintaining camera trap sampling arrays across wide geographic areas. One solution to this problem is to enlist or recruit the public for help as 'citizen scientists' collecting and processing data. Artificial Intelligence (AI) is also being used to identify animals in digital photos and video; however, this process is relatively new, and machine-based classifications are not yet fully reliable. By combining citizen science with AI, it should be possible to improve efficiency and increase classification accuracy, while simultaneously maintaining and promoting the benefits associated with public engagement with, and awareness of, wildlife.Abstract: Camera trapping has become an increasingly reliable and mainstream tool for surveying a diversity of wildlife species. Concurrent with this has been an increasing effort to involve the wider public in the research process, in an approach known as 'citizen science'. To date, millions of people have contributed to research across a wide variety of disciplines as a result. Although their value for public engagement was recognised early on, camera traps were initially ill-suited for citizen science. As camera trap technology has evolved, cameras have become more user-friendly and the enormous quantities of data they now collect has led researchers to seek assistance in classifying footage. This has now made camera trap research a prime candidate for citizen science, as reflected by the large number of camera trap projects now integrating public participation. Researchers are also turning to Artificial Intelligence (AI) to assist with classification of footage. Although this rapidly-advancing field is already proving a useful tool, accuracy is variable and AI does not provide the social and engagement benefits associated with citizen science approaches. We propose, as a solution, more efforts to combine citizen science with AI to improve classification accuracy and efficiency while maintaining public involvement.

show abstract

Recognition in Terra Incognita

Cited by 400 publications

References 55 publications

GeoStyle: Discovering Fashion Trends and Events

GeoStyle: Discovering Fashion Trends and Events

REPAIR: Removing Representation Bias by Dataset Resampling

Innovations in Camera Trapping Technology and Approaches: The Integration of Citizen Science and Artificial Intelligence

Contact Info

Product

Resources

About