Building a bird recognition app and large scale dataset with citizen scientists: The fine print in fine-grained dataset collection

Horn, Grant Van; Branson, Steve; Farrell, Ryan; Haber, Scott; Barry, Jessie H.; Ipeirotis, Panagiotis G.; Perona, Pietro; Belongie, Serge

doi:10.1109/cvpr.2015.7298658

Cited by 431 publications

(276 citation statements)

References 23 publications

Supporting

Mentioning

273

Contrasting

Order By: Relevance

“…In this work another citizen annotated dataset was used and the results show robustness to data corruption. Hence, these results confirm the discussion of [19] and motivate the use of citizen science. How much impact data corruption has in training, evaluation, and how big the dataset has to be to deal with this annotation errors, are important questions that must be answered.…”

Section: Discussionsupporting

confidence: 85%

See 1 more Smart Citation

Towards automatic wild animal monitoring: Identification of animal species in camera-trap images using very deep convolutional neural networks

Villa

Salazar

Vargas

2017

Ecological Informatics

222

103

View full text Add to dashboard Cite

Abstract. Non intrusive monitoring of animals in the wild is possible using camera trapping framework, which uses cameras triggered by sensors to take a burst of images of animals in their habitat. However camera trapping framework produces a high volume of data (in the order on thousands or millions of images), which must be analyzed by a human expert. In this work, a method for animal species identification in the wild using very deep convolutional neural networks is presented. Multiple versions of the Snapshot Serengeti dataset were used in order to probe the ability of the method to cope with different challenges that camera-trap images demand. The method reached 88.9% of accuracy in Top-1 and 98.1% in Top-5 in the evaluation set using a residual network topology. Also, the results show that the proposed method outperforms previous approximations and proves that recognition in camera-trap images can be automated.

show abstract

Section: Discussionsupporting

confidence: 85%

“…Although the comparison between experts and amateurs gives high accuracy, some mistakes occurred. In [19] it is discussed how learning algorithms are robust to annotation errors and training data corruption. This assumption is true only if the training set has sufficient samples.…”

Section: Discussionmentioning

confidence: 99%

Towards automatic wild animal monitoring: Identification of animal species in camera-trap images using very deep convolutional neural networks

Villa

Salazar

Vargas

2017

Ecological Informatics

222

103

View full text Add to dashboard Cite

show abstract

“…The distinction of finegrained categories has been studied deeply in the past [4,19,15,86,64] where applications range from fashion style recognition [30] or cars [33] to more biodiversity-driven scenarios like recognition of flowers [38], birds [82,80], dogs [29] or moths [63]. One of the most recent and promising developments is the guidance of attention to identify meaningful parts of objects [43,68] refined by advanced pooling approaches [41,69].…”

Section: Automated Detection and Identificationmentioning

confidence: 99%

Towards Automated Visual Monitoring of Individual Gorillas in the Wild

Brust

Burghardt

Groenenberg

et al. 2017

2017 IEEE International Conference on Computer Vision Workshops (ICCVW)

View full text Add to dashboard Cite

In this paper we report on the context and evaluation of a system for an automatic interpretation of sightings of individual western lowland gorillas (Gorilla gorilla gorilla) as captured in facial fieldVisual data acquisition in the field often captures sufficient information to establish encounters with individual gorillas. However, relevant information is locked within the pixel patterns measured, usually requiring expert knowledge and time-consuming efforts for identification. Computer vision can help to extract gorilla identities by performing automated species detection, followed by individual facial identification. We show that standard deep learning models combined with a traditional SVM classifier can be used for this task. To assist encounter processing, predictions can be presented graphically with known population information as shown.

show abstract

“…However, any human with good eyesight is an expert in the task of detecting a person or a dog in a picture. We found that the workforce in plentiful supply on AMT holds value even for fine grained visual categorization -the study mentioned above [14] found that expert and non-expert annotators perform roughly equally well in part localization tasks (e.g., "click on the eye"). This is good news since training a machine to identify these landmarks is a key step for species classification ( Section 2.1 ).…”

Section: Harvesting Annotations From Experts and Non-expertsmentioning

confidence: 84%

“…The answer is, unfortunately, 'no.' A recent study [14] reveals that the fine grained categories in CUB-200 [15] and ImageNet [16] , both of which used AMT to clean the datasets, have significant type I and type II errors. While as computer vision researchers we aspire to develop classification models that are robust to such errors, in the context of Visipedia image labels that are 'more or less correct' will not suffice.…”

Section: Harvesting Annotations From Experts and Non-expertsmentioning

confidence: 99%