Generalized ODIN: Detecting Out-of-Distribution Image Without Learning From Out-of-Distribution Data

Hsu, Yen-Chang; Shen, Yilin; Jin, Hongxia; Kira, Zsolt

doi:10.1109/cvpr42600.2020.01096

Cited by 343 publications

(275 citation statements)

References 9 publications

Supporting

Mentioning

275

Contrasting

Order By: Relevance

“…The novel, or unknown, species detection method in this work is important both to mosquito identification and to the field of computer vision. Previous work in recognizing novel classes, such as Outlier detection in neural networks (ODIN) and others described in the review by Geng et al on open set recognition methods, has been focused on novel classes far outside the distribution of known classes 27 . However, these algorithms do not perform well on extremely fine-grained tasks, such as that presented in this work.…”

Section: Discussionmentioning

confidence: 99%

“…Given this, a close-to-complete dataset of all mosquitoes is unlikely and any computer vision solution must be able to recognize when it encounters a species not represented in the closed-set of species as previously unencountered, rather than forcing incorrect species assignment: an open set recognition, or novelty detection, problem. The novelty detection and rejection problem has been addressed in previous computer vision work, though most successful methods, are applied to non-fine-grained problems represented by ImageNet 23 , CIFAR 24 , and MNIST 25 datasets 26 , such as the generalized ODIN of Hsu et al 27 and multi-task learning for open set recognition of Oza et al 28 .…”

Section: Main Body Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Convolutional Neural Networks That Identify Mosquito Species and Know When They Do Not Know

Goodwin¹,

Padmanabhan²,

Hira³

et al. 2021

Preprint

View full text Add to dashboard Cite

With over 3,500 mosquito species described, accurate species identification of the few implicated in disease transmission is critical to mosquito borne disease mitigation. Yet this task is hindered by limited global taxonomic expertise and specimen damage consistent across common capture methods. Convolutional neural networks (CNNs) are promising with limited sets of species, but image database requirements restrict practical implementation. Using an image database of 2,696 specimens from 67 mosquito species, we address the practical open-set problem with a detection algorithm for novel-species. Closed-set classification of 16 known species achieved 97.04±0.87% accuracy independently, and 89.07±5.58% when cascaded with novelty detection. Closed-set classification of 39 species produces a macro F1-score of 86.07±1.81%. This demonstrates an accurate, scalable, and practical computer vision solution to identify wild-caught mosquitoes for implementation in biosurveillance and targeted vector control programs, without the need for extensive image database development for each new target region.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Main Body Introductionmentioning

confidence: 99%

Convolutional Neural Networks That Identify Mosquito Species and Know When They Do Not Know

Goodwin¹,

Padmanabhan²,

Hira³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…detecting sample of low density. Many methods fall into this category, relying on the availability of ID data, possibly slightly polluted with OOD ones [1,7,12,15,17,23,30,33,36,37].…”

Section: Related Workmentioning

confidence: 99%

Sample-free white-box out-of-distribution detection for deep learning

Begon

Geurts

2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

Being able to detect irrelevant test examples with respect to deployed deep learning models is paramount to properly and safely using them. In this paper, we address the problem of rejecting such out-of-distribution (OOD) samples in a fully sample-free way, i.e., without requiring any access to indistribution or OOD samples. We propose several indicators which can be computed alongside the prediction with little additional cost, assuming white-box access to the network. These indicators prove useful, stable and complementary for OOD detection on frequently-used architectures. We also introduce a surprisingly simple, yet effective summary OOD indicator. This indicator is shown to perform well across several networks and datasets and can furthermore be easily tuned as soon as samples become available. Lastly, we discuss how to exploit this summary in real-world settings.

show abstract

“…Nowadays, target detection has made a new development. Hsu et al [12] proposed two strategies to enable the detector to detect OOD (out-of-distribution) samples without OOD data training. Wang et al [13] introduced the intersection of the human body and object into training to improve the detection performance.…”

Section: Related Workmentioning

confidence: 99%

Towards Pedestrian Target Detection with Optimized Mask R-CNN

2020

View full text Add to dashboard Cite

Aiming at the problem of low pedestrian target detection accuracy, we propose a detection algorithm based on optimized Mask R-CNN which uses the latest research results of deep learning to improve the accuracy and speed of detection results. Due to the influence of illumination, posture, background, and other factors on the human target in the natural scene image, the complexity of target information is high. SKNet is used to replace the part of the convolution module in the depth residual network model in order to extract features better so that the model can adaptively select the best convolution kernel during training. In addition, according to the statistical law, the length-width ratio of the anchor box is modified to make it more accord with the natural characteristics of the pedestrian target. Finally, a pedestrian target dataset is established by selecting suitable pedestrian images in the COCO dataset and expanded by adding noise and median filtering. The optimized algorithm is compared with the original algorithm and several other mainstream target detection algorithms on the dataset; the experimental results show that the detection accuracy and detection speed of the optimized algorithm are improved, and its detection accuracy is better than other mainstream target detection algorithms.

show abstract

Generalized ODIN: Detecting Out-of-Distribution Image Without Learning From Out-of-Distribution Data

Cited by 343 publications

References 9 publications

Convolutional Neural Networks That Identify Mosquito Species and Know When They Do Not Know

Convolutional Neural Networks That Identify Mosquito Species and Know When They Do Not Know

Sample-free white-box out-of-distribution detection for deep learning

Towards Pedestrian Target Detection with Optimized Mask R-CNN

Contact Info

Product

Resources

About