A5029857501- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Video Analysis and Summarization
- Human Pose and Action Recognition
- Soil and Unsaturated Flow
- Artificial Intelligence in Games
- Flood Risk Assessment and Management
- Digital Games and Media
- Precipitation Measurement and Analysis
- Multimodal Machine Learning Applications
- Soil Moisture and Remote Sensing
- Image Enhancement Techniques
- Generative Adversarial Networks and Image Synthesis
- AI in cancer detection
- Satellite Image Processing and Photogrammetry
- Digital Media Forensic Detection
- Agricultural Systems and Practices
- Remote Sensing and Land Use
- Landslides and related hazards
Google (United States)
2023
International Institute of Information Technology, Hyderabad
2018-2019
Indian Institute of Technology Hyderabad
2018
Abstract We develop a deep learning–based convolutional-regression model that estimates the volumetric soil moisture content in top ∼5 cm of soil. Input predictors include Sentinel-1 (active radar) and Sentinel-2 (multispectral imagery), as well geophysical variables from SoilGrids modeled fields SMAP GLDAS. The was trained evaluated on data ∼1000 situ sensors globally over period 2015–21 obtained an average per-sensor correlation 0.707 ubRMSE 0.055 m 3 −3 , it can be used to produce map at...
Abstract The increasing intensity and frequency of floods is one the many consequences our changing climate. In this work, we explore ML techniques that improve flood detection module an operational early warning system. Our method exploits unlabeled dataset paired multi-spectral synthetic aperture radar (SAR) imagery to reduce labeling requirements a purely supervised learning method. Prior works have used data by creating weak labels out them. However, from experiments, noticed such model...
Binarization is an extreme network compression approach that provides large computational speedups along with energy and memory savings, albeit at significant accuracy costs. We investigate the question of where to binarize inputs layer-level granularity show selectively binarizing specific layers in could lead improvements while preserving most advantages binarization. analyze binarization tradeoff using a metric jointly models input binarization-error cost introduce efficient algorithm...
Deep neural networks are highly effective at a range of computational tasks. However, they tend to be computationally expensive, especially in vision-related problems, and also have large memory requirements. One the most methods achieve significant improvements computational/spatial efficiency is binarize weights activations network. naive binarization results accuracy drops when applied for In this work, we present generalized, distribution-aware approach binarizing deep that allows us...
We develop a deep learning based convolutional-regression model that estimates the volumetric soil moisture content in top ~5 cm of soil. Input predictors include Sentinel-1 (active radar), Sentinel-2 (optical imagery), and SMAP (passive radar) as well geophysical variables from SoilGrids modelled fields GLDAS. The was trained evaluated on data ~1300 in-situ sensors globally over period 2015 - 2021 obtained an average per-sensor correlation 0.727 ubRMSE 0.054, can be used to produce map at...
The transition to renewable energy, particularly solar, is key mitigating climate change. Google's Solar API aids this by estimating solar potential from aerial imagery, but its impact constrained geographical coverage. This paper proposes expanding the API's reach using satellite enabling global assessment. We tackle challenges involved in building a Digital Surface Model (DSM) and roof instance segmentation lower resolution single oblique views deep learning models. Our models, trained on...
Agricultural landscapes are quite complex, especially in the Global South where fields smaller, and agricultural practices more varied. In this paper we report on our progress digitizing landscape (natural man-made) study region of India. We use high resolution imagery a UNet style segmentation model to generate first its kind national-scale multi-class panoptic output. Through work have been able identify individual across 151.7M hectares, delineating key features such as water resources...
The increasing intensity and frequency of floods is one the many consequences our changing climate. In this work, we explore ML techniques that improve flood detection module an operational early warning system. Our method exploits unlabelled dataset paired multi-spectral Synthetic Aperture Radar (SAR) imagery to reduce labeling requirements a purely supervised learning method. Prior works have used data by creating weak labels out them. However, from experiments noticed such model still...
Deep neural networks are highly effective at a range of computational tasks. However, they tend to be computationally expensive, especially in vision-related problems, and also have large memory requirements. One the most methods achieve significant improvements computational/spatial efficiency is binarize weights activations network. naive binarization results accuracy drops when applied for In this work, we present generalized, distribution-aware approach binarizing deep that allows us...
Binarization is an extreme network compression approach that provides large computational speedups along with energy and memory savings, albeit at significant accuracy costs. We investigate the question of where to binarize inputs layer-level granularity show selectively binarizing specific layers in could lead improvements while preserving most advantages binarization. analyze binarization tradeoff using a metric jointly models input binarization-error cost introduce efficient algorithm...
Video games have become an integral part of most people's lives in recent times. This led to abundance data related video being shared online. However, this comes with issues such as incorrect ratings, reviews or anything that is shared. Recommendation systems are powerful tools help users by providing them meaningful recommendations. A straightforward approach would be predict the scores based on other information game. It could used a means validate user-submitted ratings well provide work...