Artificial Intelligence

The provided data package contains Livox LiDAR scan data from the Calar simulation platform, capturing scenarios in three distinct tunnels, each with various obstacles. The dataset includes LiDAR data recorded at different distances from the sensor, . Additionally, the data encompasses varying numbers of obstacles.


The identification of rock fractures in strata is crucial to enhance the intelligence of rock detection. Traditional fracture feature extraction methods suffer from issues such as low accuracy and low processing speed, necessitating the development of more effective approaches. To address this problem, this study proposes a new fracture instance segmentation network called FracSeg. Based on the SOLOv2 framework, we incorporated the Swin Transformer to optimize the backbone network and enhance fracture feature extraction.


This dataset consists of inertial, force, color, and LiDAR data collected from a novel sensor system. The system comprises three Inertial Measurement Units (IMUs) positioned on the waist and atop each foot, a color sensor on each outer foot, a LiDAR on the back of each shank, and a custom Force-Sensing Resistor (FSR) insole featuring 13 FSRs in each shoe. 20 participants wore this sensor system whilst performing 38 combinations of 11 activities on 9 different terrains, totaling over 7.8 hours of data.


The training trajectory datasets are collected from real users when exploring the volume dataset on our interactive 3D visualization framework. The format of the training dataset collected is trajectories of POVs in the Cartesian space. Multiple volume datasets with distinct spatial features and transfer functions are used to collect comprehensive training datasets of trajectories. The initial point is randomly selected for each user. Collected training trajectories are cleaned by removing POV outliers due to users' misoperations to improve uniformity.


This study is based on the image data of cement concrete pavement diseases collected by myself. The mobile phone is fixed on the sun visor of the passenger seat of the vehicle, and all kinds of diseases on the road are photographed along with the vehicle. Based on 1,595 images, each image is expanded to 4 by using the data enhancement method. After screening, a total of 2,925 images are obtained, including 2,125 defective images with shadow occlusion and uneven illumination.


Subjects are categorized into three groups based on office blood pressure threshold: Normal (N), Prehypertension (P), and Stage 1 Hypertension (S). Each group contains 100 subjects, and all records have duration of at least 8 minutes. This study uses sliding window with length of 1 second and step size of 1 second to segment records. PPG, ECG and BP yield 167432 segments, respectively. MAP, DBP, and SBP are defined as average, minimum, and maximum of each BP segment, respectively. Max-Min normalization is applied to PPG and ECG segments. 


This dataset comprises three benchmarks: Digits-5, PACS, anf office_caltech_10. Digits-5 is a set of handwritten digit images sampled from five domains: MNIST, MNIST-M, USPS, SynthDigits, and SVHN.  All sample are images of numbers ranging from 0 to 9.  PACS is composed of four different datasets, each representing a different visual domain: Photo, Art Painting, Cartoon, and Sketch. It contains 9,944 images, including 1,792 real photos, 2,048 art paintings, 2,344 cartoon images, and 2,760 sketches.


Bengaluru has been ranked the most congested city in India in terms of traffic for several years now. This hackathon is aimed at creating innovative solutions to the traffic management problem in Bengaluru, and is co-sponsored by the Bengaluru Traffic Police, the Centre for Data for Public Good, and the Indian Institute of Science (IISc). The hackathon will have two phases. The first phase will be about short-term traffic volume prediction, given video feeds from cameras installed at junctions.

Last Updated On: 
Fri, 07/19/2024 - 07:51
Citation Author(s): 
Raghu Krishnapuram, Rakshit Ramesh, and Arun Josephraj

Dynamic nonlinear equations (DNEs) are essential for modeling complex systems in various fields due to their ability to capture real-world phenomena. However, the solution of DNEs presents significant challenges, especially in industrial settings where periodic noise often compromises solution fidelity. To tackle this challenge, we propose a novel approach called Periodic Noise Suppression Neural Dynamic (PNSND), which leverages the gradient descent approach and incorporates velocity compensation to overcome the limitations of the traditional Gradient Neural Dynamic (GND) model.


This paper presents a comparative study of sampling methods within the FedHome framework, designed for personalized in-home health monitoring. FedHome leverages federated learning (FL) and generative convolutional autoencoders (GCAE) to train models on decentralized edge devices while prioritizing data privacy. A notable challenge in this domain is the class imbalance in health data, where critical events such as falls are underrepresented, adversely affecting model performance.