| 1. |
- Gustafsson, Fredrik K., et al.
(författare)
-
Evaluating Scalable Bayesian Deep Learning Methods for Robust Computer Vision
- 2020
-
Ingår i: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW 2020). - : IEEE Computer Society. - 9781728193601 ; , s. 1289-1298
-
Konferensbidrag (refereegranskat)abstract
- While deep neural networks have become the go-to approach in computer vision, the vast majority of these models fail to properly capture the uncertainty inherent in their predictions. Estimating this predictive uncertainty can be crucial, for example in automotive applications. In Bayesian deep learning, predictive uncertainty is commonly decomposed into the distinct types of aleatoric and epistemic uncertainty. The former can be estimated by letting a neural network output the parameters of a certain probability distribution. Epistemic uncertainty estimation is a more challenging problem, and while different scalable methods recently have emerged, no extensive comparison has been performed in a real-world setting. We therefore accept this task and propose a comprehensive evaluation framework for scalable epistemic uncertainty estimation methods in deep learning. Our proposed framework is specifically designed to test the robustness required in real-world computer vision applications. We also apply this framework to provide the first properly extensive and conclusive comparison of the two current state-of-the-art scalable methods: ensembling and MC-dropout. Our comparison demonstrates that ensembling consistently provides more reliable and practically useful uncertainty estimates. Code is available at https://github.com/fregu856/evaluating_bdl.
|
|
| 2. |
- Nakasi, Rose, et al.
(författare)
-
A web-based intelligence platform for diagnosis of malaria in thick blood smear images : A case for a developing country
- 2020
-
Ingår i: Proceedings - 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2020. - 2160-7516 .- 2160-7508. - 9781728193601 ; 2020-June, s. 4238-4244
-
Konferensbidrag (refereegranskat)abstract
- Malaria is a public health problem which affects developing countries world-wide. Inadequate skilled lab technicians in remote areas of developing countries result in untimely diagnosis of malaria parasites making it hard for effective control of the disease in highly endemic areas. The development of remote systems that can provide fast, accurate and timely diagnosis is thus a necessary innovation. With availability of internet, mobile phones and computers, rapid dissemination and timely reporting of medical image analytics is possible. This study aimed at developing and implementing an automated web-based Malaria diagnostic system for thick blood smear images under light microscopy to identify parasites. We implement an image processing algorithm based on a pre-trained model of Faster Convolutional Neural Network (Faster R-CNN) and then integrate it with web-based technology to allow easy and convenient online identification of parasites by medical practitioners. Experiments carried out on the online system with test images showed that the system could identify pathogens with a mean average precision of 0.9306. The system holds the potential to improve the efficiency and accuracy in malaria diagnosis, especially in remote areas of developing countries that lack adequate skilled labor.
|
|
| 3. |
- Ornhag, Marcus Valtonen, et al.
(författare)
-
Bilinear parameterization for differentiable rank-regularization
- 2020
-
Ingår i: IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops. - 2160-7516 .- 2160-7508. - 9781728193601 ; 2020-June, s. 1416-1425
-
Konferensbidrag (refereegranskat)abstract
- Low rank approximation is a commonly occurring problem in many computer vision and machine learning applications. There are two common ways of optimizing the resulting models. Either the set of matrices with a given rank can be explicitly parametrized using a bilinear factorization, or low rank can be implicitly enforced using regularization terms penalizing non-zero singular values. While the former approach results in differentiable problems that can be efficiently optimized using local quadratic approximation, the latter is typically not differentiable (sometimes even discontinuous) and requires first order subgradient or splitting methods. It is well known that gradient based methods exhibit slow convergence for ill-conditioned problems.In this paper we show how many non-differentiable regularization methods can be reformulated into smooth objectives using bilinear parameterization. This allows us to use standard second order methods, such as Levenberg- Marquardt (LM) and Variable Projection (VarPro), to achieve accurate solutions for ill-conditioned cases. We show on several real and synthetic experiments that our second order formulation converges to substantially more accurate solutions than competing state-of-the-art methods.1
|
|
