Papers by Praveen Narayanan
2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021
A distinctive feature of Doppler radar is the measurement of velocity in the radial direction for... more A distinctive feature of Doppler radar is the measurement of velocity in the radial direction for radar points. However, the missing tangential velocity component hampers object velocity estimation as well as temporal integration of radar sweeps in dynamic scenes. Recognizing that fusing camera with radar provides complementary information to radar, in this paper we present a closed-form solution for the point-wise, full-velocity estimate of Doppler returns using the corresponding optical flow from camera images. Additionally, we address the association problem between radar returns and camera images with a neural network that is trained to estimate radar-camera correspondences. Experimental results on the nuScenes dataset verify the validity of the method and show significant improvements over the state-of-the-art in velocity estimation and accumulation of radar points.
Direct numerical Simulation of turbulent nonpremixed flame
2019 International Conference on Robotics and Automation (ICRA), 2019
We present a Deep Learning based system for the twin tasks of localization and obstacle avoidance... more We present a Deep Learning based system for the twin tasks of localization and obstacle avoidance essential to any mobile robot. Our system learns from conventional geometric SLAM, and outputs, using a single camera, the topological pose of the camera in an environment, and the depth map of obstacles around it. We use a CNN to localize in a topological map, and a conditional VAE to output depth for a camera image, conditional on this topological location estimation. We demonstrate the effectiveness of our monocular localization and depth estimation system on simulated and real datasets.
We present a voice conversion solution using recurrent sequence to sequence modeling for DNNs. Ou... more We present a voice conversion solution using recurrent sequence to sequence modeling for DNNs. Our solution takes advantage of recent advances in attention based modeling in the fields of Neural Machine Translation (NMT), Text-to-Speech (TTS) and Automatic Speech Recognition (ASR). The problem consists of converting between voices in a parallel setting when {\it $ $} audio pairs are available. Our seq2seq architecture makes use of a hierarchical encoder to summarize input audio frames. On the decoder side, we use an attention based architecture used in recent TTS works. Since there is a dearth of large multispeaker voice conversion databases needed for training DNNs, we resort to training the network with a large single speaker dataset as an autoencoder. This is then adapted for the smaller multispeaker voice conversion datasets available for voice conversion. In contrast with other voice conversion works that use $F_0$, duration and linguistic features, our system uses mel spectrog...
ArXiv, 2020
Generative Adversarial Networks (GANs) are now widely used for photo-realistic image synthesis. I... more Generative Adversarial Networks (GANs) are now widely used for photo-realistic image synthesis. In applications where a simulated image needs to be translated into a realistic image (sim-to-real), GANs trained on unpaired data from the two domains are susceptible to failure in semantic content retention as the image is translated from one domain to the other. This failure mode is more pronounced in cases where the real data lacks content diversity, resulting in a content \emph{mismatch} between the two domains - a situation often encountered in real-world deployment. In this paper, we investigate the role of the discriminator's receptive field in GANs for unsupervised image-to-image translation with mismatched data, and study its effect on semantic content retention. Experiments with the discriminator architecture of a state-of-the-art coupled Variational Auto-Encoder (VAE) - GAN model on diverse, mismatched datasets show that the discriminator receptive field is directly correl...
Fire Safety Science, 2011
The objective of the present study is to use large activation energy asymptotic (AEA) theory to b... more The objective of the present study is to use large activation energy asymptotic (AEA) theory to bring basic information on the extinction limits of non-premixed flames. The AEA analysis leads to an explicit expression that predicts the occurrence of flame extinction in the form of a critical Damköhler number criterion; the criterion provides a unified framework to explain the different extinction limits that are observed in non-premixed combustion (i.e., aerodynamic quenching, thermal quenching, and dilution quenching). The critical Damköhler number criterion is then formulated in terms of six input variables; these variables characterize the magnitude of flame stretch, the magnitude of the flame heat losses, and the composition and heat content of the fuel and oxidizer supply streams; these input variables thereby contain information on (laminar or turbulent) flow-induced perturbations, deviations from adiabatic combustion, and air and fuel vitiation. Different two-dimensional flammability maps are then presented using different assumptions aimed at reducing the dimension of the parameter space from six to two. While providing a limited view point, these flammability maps provide valuable insights; it is found for instance that diffusion flames are more sensitive to air vitiation than fuel vitiation.
Radiation-driven flame weakening effects in sooting turbulent diffusion flames
Proceedings of the Combustion Institute, 2009
ABSTRACT
Effect of soot addition on extinction limits of luminous laminar counterflow diffusion flames
Proceedings of the Combustion Institute, 2011
ABSTRACT The objective of the present study is to use large activation energy asymptotic (AEA) th... more ABSTRACT The objective of the present study is to use large activation energy asymptotic (AEA) theory to bring basic information on the extinction limits of non-premixed flames under sooting and radiating conditions. The AEA analysis assumes single-step global combustion chemistry, constant heat capacity and unity Lewis numbers; it also includes a two-equation phenomenological model to describe soot formation, growth and oxidation processes, as well as a generalized treatment of thermal radiation that assumes spectrally-averaged gray-medium properties and applies to flames with an arbitrary optical thickness. The focus of the present study is on the effect of external soot loading on flame extinction, and in particular on the slow-mixing/radiative-extinction limit that is believed to be the dominant mechanism that determines flame extinction in fires. External soot loading simulates non-local effects observed in multi-dimensional sooting flames in which soot mass may be produced at some flame locations and transported to others where it will increase the flame luminosity and drive combustion conditions towards extinction. The AEA analysis shows that external soot loading results in a significant decrease of the size of the flammable domain and that the minimum value of flame stretch at the radiative-extinction limit is increased by more than one order of magnitude compared to a non-soot-loaded-flame case. Multi-dimensional sooting flames are therefore expected to be significantly more susceptible to radiative extinction than the one-dimensional configurations that have been previously studied in microgravity combustion research.
Proceedings of the Combustion Institute, 2011
The interaction of turbulent non-premixed flames with fine water spray is studied using direct nu... more The interaction of turbulent non-premixed flames with fine water spray is studied using direct numerical simulations (DNS) with detailed chemistry. The study is of practical importance in fire safety devices that operate in the mist regime, as well as in their use as an inexpensive temperature control mechanism for gas turbines. Dynamics of water spray is represented by the Lagrangian particle-in-cell method, coupled with an Eulerian gas-phase reacting flow solver. The model configuration is a two-dimensional ethylene-air counterflow diffusion flame at moderate strain rates. Laminar and turbulent flame simulations are performed at various water loading conditions. Comparison of various simulation cases highlights the flame weakening characteristics due to aerodynamic stretch and heat loss due to water spray evaporation. Local flame extinction is identified by a flame weakness factor derived based on an asymptotic model under nonadiabatic environments. A statistical analysis of the cumulative turbulent flame data shows that a large heat release enhancement is observed during the flame quenching due to the occurrence of edge flames, while such effects are substantially reduced in the presence of water spray. Findings from this study provide a better understanding of interaction between thermal and aerodynamic quenching in turbulent flame dynamics.
Journal of Physics: Conference Series, 2007
This paper presents our recent progress in terascale three-dimensional simulations of turbulent n... more This paper presents our recent progress in terascale three-dimensional simulations of turbulent nonpremixed flames in the presence of a mean flow strain and fine water droplets. Under the ongoing university collaborative project supported by the DOE SciDAC Program [1] along with the INCITE 2007 Project [2], the study aims at bringing the state-of-the-art highfidelity simulation capability to the next level by incorporating various advanced physical models for soot formation, radiative heat transfer, and lagrangian spray dynamics, to an unprecedented degree of detail in high-fidelity simulation application. The targeted science issue is fundamental characteristics of flame suppression by the complex interaction between turbulence, chemistry, radiation, and water spray. The high quality simulation data with full consideration of multi-physics processes will allow fundamental understanding of the key physical and chemical mechanisms in the flame quenching behavior. In this paper, recent efforts on numerical algorithms and model development toward the targeted terascale 3D simulations are discussed and some preliminary results are presented.
ftg.lbl.gov
Magnetic fusion is a long-term solution for producing electrical power for the world, and the lar... more Magnetic fusion is a long-term solution for producing electrical power for the world, and the large thermonuclear international device (ITER) being constructed will produce net energy and a path to fusion energy provided the computer modeling is accurate. To effectively address the requirements of the high-end fusion simulation community, application developers, algorithm designers, and hardware architects must have reliable simulation data gathered at scale for scientifically valid configurations. This paper presents detailed benchmarking results for a set of magnetic fusion applications with a wide variety of underlying mathematical models including both particle-in-cell and Eulerian codes using both implicit and explicit numerical solvers. Our evaluation on a petascale Cray XE6 platform focuses on profiling these simulations at scale identifying critical performance characteristics, including scalability, memory/network bandwidth limitations, and communication overhead. Overall results are a key in improving fusion code design, and are a critical first step towards exascale hardware-software co-design-a process that tightly couples applications, algorithms, implementation, and computer architecture.
Edge Localized Modes Fast (∼ 100µs) eruption from the edge of tokamak plasmas If uncontrolled in ... more Edge Localized Modes Fast (∼ 100µs) eruption from the edge of tokamak plasmas If uncontrolled in ITER, these would release ∼ 20 MJ Worldwide effort to understand and control these events The BOUT++ Simulation Code Based on BOUT written by X. Xu, et. al. from LLNL [1] New 3D simulation code developed at York with LLNL and ANL Simulates plasma fluid equations in curvilinear coordinate systems Runs on workstations, clusters, large-scale machines, e.g., Cray XE6
2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021
Figure 1: Radar-camera depth completion: (a) an image with 0.3 seconds (5 sweeps) of radar hits p... more Figure 1: Radar-camera depth completion: (a) an image with 0.3 seconds (5 sweeps) of radar hits projected onto it, (b) enhanced radar depths at confidence level 0.9 eliminate occluded pixels and expand visible hits, and (c) final predicted depth through depth completion.
Probabilistic Inference by Program Transformation in Hakaru (System Description)
Lecture Notes in Computer Science, 2016
Journal of Physics: Conference Series, 2005
ØEngine applications: Flame-wall interactions (FWI) are observed in many combustion systems (i.e.... more ØEngine applications: Flame-wall interactions (FWI) are observed in many combustion systems (i.e. IC engines, micro-scale combustors); FWI result in decreased performance, both in terms of thermal efficiency and pollution propensity; ØFire applications: FWI are also observed in many fire configurations, whenever the flame develops near a solid (inert/flammable) surface (i.e. wall fires, impinging flames); FWI contribute to the wall surface heat transfer (i.e. to the fuel pyrolysis) and to the thermal loading experienced by solid structures; ØCFD modeling: FWI are simply neglected in current engineering-level CFD models (wall boundary layer models, developed for inert flows, are inadequate).
Uploads
Papers by Praveen Narayanan