Learning to Downsample for Segmentation of Ultra-High Resolution Images

Chen, Jin; Tanno, Ryutaro; Mertzanidou, Thomy; Panagiotaki, Eleftheria; Alexander, Daniel C.

doi:10.48550/arxiv.2109.11071

Cited by 3 publications

(4 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, when we use the segmented blood vessels, the retinal blood vessels are extracted, while the colored image is converted to greyscale images, and the region outside the retinal blood vessel is filtered out. As a result of down-sampling in the segmentation process, some information (pixel values) from the retinal blood vessels is lost due to segmentation [ 48 ]. In addition, the region outside the vessel can contain important information (such as drusen and other biomarkers) which are essential biomarkers in a glaucoma diagnosis.…”

Section: Discussionmentioning

confidence: 99%

Automated Diagnosis of Diabetic Retinopathy Using Deep Learning: On the Search of Segmented Retinal Blood Vessel Images for Better Performance

et al. 2023

View full text Add to dashboard Cite

Diabetic retinopathy is one of the most significant retinal diseases that can lead to blindness. As a result, it is critical to receive a prompt diagnosis of the disease. Manual screening can result in misdiagnosis due to human error and limited human capability. In such cases, using a deep learning-based automated diagnosis of the disease could aid in early detection and treatment. In deep learning-based analysis, the original and segmented blood vessels are typically used for diagnosis. However, it is still unclear which approach is superior. In this study, a comparison of two deep learning approaches (Inception v3 and DenseNet-121) was performed on two different datasets of colored images and segmented images. The study’s findings revealed that the accuracy for original images on both Inception v3 and DenseNet-121 equaled 0.8 or higher, whereas the segmented retinal blood vessels under both approaches provided an accuracy of just greater than 0.6, demonstrating that the segmented vessels do not add much utility to the deep learning-based analysis. The study’s findings show that the original-colored images are more significant in diagnosing retinopathy than the extracted retinal blood vessels.

show abstract

Section: Discussionmentioning

confidence: 99%

Automated Diagnosis of Diabetic Retinopathy Using Deep Learning: On the Search of Segmented Retinal Blood Vessel Images for Better Performance

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Similarly, the method in [57] utilizes non-uniform downsampling for semantic segmentation. However, in contrast with the previous method, the saliency detector in this method is optimized based on the performance of semantic segmentation rather than external supervision signals.…”

Section: Methods For Efficient Processing Of High-resolution Inputs W...mentioning

confidence: 99%

Efficient High-Resolution Deep Learning: A Survey

Bakhtiarnia¹,

Zhang²,

Iosifidis³

2022

Preprint

View full text Add to dashboard Cite

Cameras in modern devices such as smartphones, satellites and medical equipment are capable of capturing very high resolution images and videos. Such high-resolution data often need to be processed by deep learning models for cancer detection, automated road navigation, weather prediction, surveillance, optimizing agricultural processes and many other applications. Using high-resolution images and videos as direct inputs for deep learning models creates many challenges due to their high number of parameters, computation cost, inference latency and GPU memory consumption. Simple approaches such as resizing the images to a lower resolution are common in the literature, however, they typically significantly decrease accuracy. Several works in the literature propose better alternatives in order to deal with the challenges of high-resolution data and improve accuracy and speed while complying with hardware limitations and time restrictions. This survey describes such efficient high-resolution deep learning methods, summarizes realworld applications of high-resolution deep learning, and provides comprehensive information about available high-resolution datasets.

show abstract

“…Deformed downsampling [22] is inspired from the way humans visualize images. The idea is to identify high resolution details in patches that contain important boundary details that contribute to enhanced image segmentation and ignore the high resolution details of other patches.…”

Section: Existing Literature and Related Workmentioning

confidence: 99%

“…Several of the recent approaches [28], [3], [5], [8] involve design of neural network architectures that can combine local and global information from downsampled images and multiple cropped patches of high resolution images in various ways. Other recent approaches [23], [22] involve design of neural networks that are capable of identifying important patches in the high resolution images and are based on the idea of non-uniform sampling of the pixels to reduce the computational burden. Although these methods alleviate the memory issues for model training, most of them require ground truth annotations of the high resolution images.…”

Section: Introductionmentioning

confidence: 99%

A Robust Morphological Approach for Semantic Segmentation of Very High Resolution Images

Saravanan¹,

Challa²,

Danda³

2022

Preprint

View full text Add to dashboard Cite

State-of-the-art methods for semantic segmentation of images involve computationally intensive neural network architectures. Most of these methods are not adaptable to high-resolution image segmentation due to memory and other computational issues. Typical approaches in literature involve design of neural network architectures that can fuse global information from low-resolution images and local information from the high-resolution counterparts. However, architectures designed for processing high resolution images are unnecessarily complex and involve a lot of hyper parameters that can be difficult to tune. Also, most of these architectures require ground truth annotations of the high resolution images to train, which can be hard to obtain. In this article, we develop a robust pipeline based on mathematical morphological (MM) operators that can seamlessly extend any existing semantic segmentation algorithm to high resolution images. Our method does not require the ground truth annotations of the high resolution images. It is based on efficiently utilizing information from the lowresolution counterparts, and gradient information on the highresolution images. We obtain high quality seeds from the inferred labels on low-resolution images using traditional morphological operators and propagate seed labels using a random walker to refine the semantic labels at the boundaries. We show that the semantic segmentation results obtained by our method beat the existing state-of-the-art algorithms on high-resolution images. We empirically prove the robustness of our approach to the hyper parameters used in our pipeline. Further, we characterize some necessary conditions under which our pipeline is applicable and provide an in-depth analysis of the proposed approach.

show abstract

Learning to Downsample for Segmentation of Ultra-High Resolution Images

Cited by 3 publications

References 33 publications

Automated Diagnosis of Diabetic Retinopathy Using Deep Learning: On the Search of Segmented Retinal Blood Vessel Images for Better Performance

Automated Diagnosis of Diabetic Retinopathy Using Deep Learning: On the Search of Segmented Retinal Blood Vessel Images for Better Performance

Efficient High-Resolution Deep Learning: A Survey

A Robust Morphological Approach for Semantic Segmentation of Very High Resolution Images

Contact Info

Product

Resources

About