A higher-order differential optical-flow method using a second-order Taylor formulation, perturbation-based correction, and Horn–Schunck regularization, evaluated on KITTI and Middlebury benchmarks.

Pinpointing where poverty is most severe and tracking its changes over time is crucial for facilitating and monitoring the effects of poverty alleviation initiatives. However, traditional benchmarks like household surveys and national censuses often fall short—they’re expensive, resource-intensive, infrequent, and incapable of reflecting the full spectrum of household well-being. They often fail to account for geographic variation in cost of living, medical needs, or the costs of earning income. This gap is particularly pronounced in contexts like Sri Lanka, leading to a reliance on obsolete information when responding to economic shocks or disasters. On top of that, poverty cannot be determined by income data alone; rather, it’s multidimensional, where factors such as infrastructure, access to services, and economic activity also play a role in determining the well-being of a community

This research introduced an end-to-end open-source pipeline to predict the Built-up (Bu) and Non-Built-up (NBU) areas using satellite band data. Finally, the project provides the code base for implementing the UN- Habitat definition What is a City, to understand the patterns, trends, and boundaries of urbanization.

Primary Focus: RAG(Retrieval augmented generation) faces limitations in capturing context and understanding complex queries as the proximity of text chunks in the embedding space does not guarantee a meaningful question-and-answer pair.

This works by using a deep neural network to take two images—a content image and a feature image—and blend them together so that the output image looks like the content image but "painted" in the style of the style image. VGG-19, a 19-layer CNN trained on the ImageNet dataset, is utilized here. Since this implementation doesn't require training or fully connected layers, the trainable parameters are set to False. To obtain a multi-level representation of the style image, feature correlations are calculated between different feature maps. The content loss between the content image and the output image is computed using the mean squared error (MSE) of localized shapes. However, for the style image, calculating loss through MSE is less effective, as the goal is to mimic the color and texture in the output image. Therefore, the Gram matrix is used to capture a more general representation of the style image.