Image Processing

To evaluate SARNet’s generalization, we captured a real-world stereo dataset in Guangzhou using a binocular camera. The dataset includes diverse urban and natural scenes to assess SARNet’s performance beyond synthetic and benchmark datasets. Fig. 7 illustrates SARNet’s predictions on real-world scenes, KITTI 2012, and KITTI 2015. Experimental results demonstrate that SARNet generates clear and consistent disparity maps across both smooth and complex regions, highlighting its robustness in real-world depth estimation tasks.

The Ancient Handwritten Devanagari Documents dataset is a curated collection of historical manuscripts written in the Devanagari script. It comprises digitized images of handwritten texts from various periods, containing diverse calligraphic styles, degradations, and linguistic variations. This dataset is designed for research in optical character recognition (OCR), handwritten text recognition (HTR), word spotting in historical documents and linguistic analysis.

This dataset comprises 33,800 images of underwater signals captured in aquatic environments. Each signal is presented against three types of backgrounds: pool, marine, and plain white. Additionally, the dataset includes three water tones: clear, blue, and green. A total of 12 different signals are included, each available in all six possible background-tone combinations.

This dataset contains high-resolution retinal fundus images collected from 495 unique subjects from Eye Care hospital in Aizawl, Mizoram, for diabetic retinopathy (DR) detection and classification. The images were captured over five years using the OCT RS 330 device, which features a 45° field of view (33° for small-pupil imaging), a focal length of 45.7 mm, and a 6.25 mm sensor width. Each image was acquired at a resolution of 3000x3000 pixels, ensuring high diagnostic quality and the visibility of subtle features like microaneurysms, exudates, and hemorrhages.