Indian Institute of Science Bangalore
etd@IISc Electronic Theses and Dissertations at Indian Institute of ScienceNot a member yet
6204 research outputs found
Sort by
High Dimensional Quantum Information Technology Using Photonics: Protocol Development and Implementation Schemes
No full textThis thesis explores the realm of high-dimensional quantum information processing through the lens of photonics, focusing on developing and implementing quantum technologies based on d-level quantum systems known as qudits. Quantum information technology, with its transformative impact on information processing, has spurred innovations such as quantum computers, quantum cryptography, and quantum internet. However, the implementation challenges faced by current quantum information technology necessitate novel approaches, and photonics emerges as a promising solution.
The research begins with a foundational analysis of sources and detectors for qudits, leveraging integrated photonic devices based on light propagation in optical waveguides. The modes of optical waveguides are proposed to represent qudits, and non-linear optical processes in micro-ring resonators are explored to develop single and multi-dimensional photon sources. A scheme for generating single photons using integrated optic ring resonators is presented.
A cornerstone contribution lies in the development of a High-Dimensional version of the Quantum Key Distribution (QKD) Protocol BB84, introducing qudits into the realm of secure key distribution. The design and simulation of waveguide-based circuits for the practical implementation of both standard QKD and High-Dimensional QKD protocols support this conceptual advancement. The latter involves multi-mode waveguide-based circuits, enhancing the understanding of quantum communication in higher dimensions.
A W-state encoding scheme for photonic qudits is proposed to address the critical aspect of error correction in quantum communication. This scheme holds promise as an error correction strategy for quantum memories and communication systems operating in higher dimensions.
The thesis also extends its theoretical contributions into the experimental domain through collaborations with leading laboratories. The proposed experimental validation of the W-state protocol for qubits and qudits bridges the gap between theoretical frameworks and real-world applications.
In summary, this thesis contributes significantly to the field of High-Dimensional quantum information processing using photonics. The integration of qudits into key quantum protocols, the design of practical waveguide-based circuits, and the development of novel error correction schemes lay the groundwork for the next frontier in quantum technology. The collaboration with esteemed laboratories further enriches the research, promising advancements that transcend theoretical boundaries and manifest in practical quantum communication systems.Visvesvaraya PhD Fellowship Program, Ministry of Electronics and IT (MeitY), Govt. of India
Fracture and Fatigue Behaviour of Notched Plain Concrete Beams: The Role of Theory of Critical Distances and Acoustic Emission
No full textConcrete structures, including buildings, bridges, pavements, and offshore structures,
face a wide range of loading conditions, both static and cyclic. When subjected to
fatigue loading, the response of concrete elements can be quite complex and difficult
to predict, primarily due to the inherent heterogeneity of the material. Experiments
are the most effective approach to study fatigue behaviour, utilizing techniques like
acoustic emission to understand internal microcracking. Typically, these experiments
are performed on notched specimens to accurately monitor fracture behavior. However,
fatigue experiments are time-consuming and relatively expensive considering the casting
and curing time in addition to the cycling time during tests. To optimize costs and save
time, it is beneficial to determine the static strength of specimens prior to the experiment.
This knowledge helps in determining load amplitudes for testing and enables the effective
design of experiments.
The theory of critical distances (TCD), due to its appealing characteristics, has been
successfully used in the past to predict the strength of brittle as well as ductile materials,
weakened by the presence of stress risers, under both static and fatigue loading. In this
work, the TCD’s unique features are exploited, and the point method is reformulated
to predict the strength of notched plain concrete beams of different sizes under mode
I quasi-static loading. The presence of fracture process zone, which is responsible for
the post-peak softening behaviour of concrete under tension, is considered through the
concept of an effective elastic crack. A power law is proposed to relate the effective
crack length to the geometrical properties. The material characteristic length, required
for the application of TCD, is correlated with the maximum aggregate size. The resulting
formulation is found to yield satisfactory predictions of static strength of notched plain
concrete beams, wherein the geometric dimensions of the beam, tensile strength, and
maximum aggregate size of the concrete mix are the governing parameters. The proposed
formulation is validated using a probabilistic analysis of various experimental results
available in the literature. Furthermore, the TCD is applied for predicting the static
strength under mixed mode loading. Two alternatives are proposed, one by directly
applying TCD by considering the characteristic length to vary linearly with mode mixity
ratio and the other by converting the mixed mode problem into an equivalent mode I
case using energy equivalence.
Understanding the internal microcracking occurring within concrete is crucial to gain
insight into its behaviour under fatigue conditions. One of the most effective techniques
for achieving this is the use of acoustic emission (AE). This study explores the potential of various methods for analysing AE signals such as average frequency versus rise
angle analysis and intensity analysis for characterisation of the fracture process in plain
concrete under both monotonic and fatigue loading conditions. By applying k-means
clustering to the AE data, four damage mechanisms in plain concrete, namely cement
mortar cracking, aggregate slip, ITZ cracking, and aggregate fracture, are identified.
In addition, parameters extracted from AE signals have shown a promising correlation
with fatigue crack behaviour, with a log-linear relationship between crack propagation
rate and AE parameter rate. Among the different parameters extracted from the AE
waveforms, the AE energy is identified as the most suitable parameter for characterising
fatigue behaviour of concrete. Here, the model parameters as well as their posterior
distributions are estimated using Bayesian regression.
The present study is, thus, an attempt to understand and forecast the response of con crete specimens when subjected to monotonic and fatigue loading utilising the potential
of the theory of critical distances and acoustic emission. Overall, this dissertation aims
to provide valuable insights into the response of concrete to different loading conditions
and contribute to the development of more accurate and reliable methods for predicting
the behaviour of concrete structures under real world loading conditions
Orbital Engineering Directed Electronic Structure Studies: Design of Pseudo Aromatic Molecules, Borophenes and Borophites
No full textThe many ways of using three-membered rings as a design element in molecules and borophenes (2D sheets of boron) will be presented. The underlying design principles require constructing and manipulating molecular orbitals and band structure with chemical intuition using "Orbital Engineering" [1]. The out-of-plane (π) delocalization in an aromatic system is not confined to the overlap of standard p-, d-, and f- orbitals. We present an alternative by choosing σ* Fragment Molecular Orbitals (FMO) with π symmetry, named pseudo π* FMOs [2,3]. The all-bonding combination of these in cyclic systems creates pseudo π* aromaticity. Therefore, we utilize our orbital engineering principles to design stable pseudo π* aromatic systems, starting from the smallest aromatic ring, cyclopropenium cation (C3H3+). We stabilize a three-membered Si2B ring involving two pseudo π* FMOs based on Si and one pure p- orbital based on B, which is synthesized by the Roesky group [3]. We extend this concept to the saturated three-, four-, five- and six-membered rings of Silicon involving only pseudo π* FMOs. Polycondensation of all-boron analogs of C3H3+ such as B3H5 and B3H6+ leads to borophene sheets with varying hole density, with the three-membered B3 ring as a continuum. Based on this, we present a generalized electron counting rule for borophenes with any hole density [4]. Knowing the electronic requirement, we propose various ways to stabilize both electron-deficient and rich borophenes. In electron-deficient borophenes, such as β12 sheet with a hole density of 1/6, we suggest hydrogenation and metal doping as the ways of electron donation, detaching the 2D layer from the metal surface on which theses are generated [4,5]. On the other hand, for electron-rich borophenes, the excess electrons are localized by creating interlayer bonds between two layers [6]. Interlayer bonding can be used as a design element to construct multilayers of borophenes: the number and nature of which change based on hole density on the borophene sheets and the number of layers [7]. We also examine the potential existence of layered boron materials: borophites, similar to graphite, where van der Waals forces stack borophene layers [8,9]. However, unlike graphite, in borophites, the constituent layers can be a monolayer, a bilayer, or even a multilayer. Quantitative studies on these will be presented.DST INSPIR
Expanding the Possibilities of Multifunctional Gelatin Methacrylol Bioinks for 3D Bioprinting of Complex Biomimetic Tissue Constructs
No full textTissue engineering, or the construction of complex tissue-like structures for injured tissue replacement and regeneration, has been a subject of investigation over the last thirty years. In the last decade, there has been considerable interest in using additive manufacturing (3D printing) to achieve these goals. Despite such efforts, many key questions still need to be answered, particularly in relation to the scientific understanding of the printability and buildability of the biomaterial inks for targeted tissue-specific applications.
In addressing many unanswered questions, this dissertation reports the formulation and 3D bioprinting studies on hybridized bioinks with inorganic nanofillers based on the model baseline biomaterial, gelatin methacryloyl (GelMA), while demonstrating how they can be used as a potential scaffold for the transport of living cells as well as their maintenance for different tissue remodeling, particularly bone, cartilage, and nerve. Recognizing the importance of biophysical properties, printability toward shape fidelity, and biofunctionality, a significant focus has been centered around the qualitative and quantitative evaluation of the role of printing parameters, microrheological properties, and pore architectures on the mechanical properties, swelling kinetics, enzymatic degradation, and printability towards shape fidelity of biomaterial inks, in particular reference to 3D extrusion printing.
In order to identify the role of inorganic fillers in the scaffold buildability, varying amounts of hydrothermally synthesized phase pure rod-shaped nanocrystalline hydroxyapatite (HAp) powders were incorporated into pre-crosslinked GelMA matrix to fabricate a predesigned scaffold architecture using a custom-made 3D bioprinter. The HAp-incorporated GelMA compositions demonstrated superior printability and scaffold stability, with uniform distribution of HAp nanoparticles without any phase separation. Notably, the experimental results clearly suggested that the uniaxial compression properties, swelling behavior, and biodegradations can be regulated by optimizing the HAp content.
Further, we addressed the shortcomings of inferior printability of low-concentration GelMA by methacrylated carboxymethyl cellulose (mCMC), which has a significant impact on extrudability and buildability, sol-gel transition temperature, and yield stress. These photopolymerizable GelMA/mCMC ink served as the foundation for the growth and development of cartilage matrix after encapsulation of human mesenchymal stem cells (hMSCs). The inclusion of nHAp provided enhanced bioactivity and osteoinductive properties to promote the formation of a new bone matrix. Intriguingly, we incorporated poly(ethylene glycol)diacrylate (PEGDA) into the viscosity-modified GelMA to enhance compressive modulus and regulate cell (hMSCs) functionality. This knowledge contributed to optimizing the biomaterials ink formulation, ultimately improving the printability, buildability, and overall performance of the GelMA matrix in 3D bioprinting applications.
Next, we investigated the potential of GelMA to develop fully integrated, multilayered nerve conduits. We systematically incorporated carbon nanofiber (CNF), PEGDA, and gellan gum (GG) to synthesize an electroconductive bioink. We have semi-quantitatively analyzed the limitations of extrusion bioprinting to reconstruct freestanding thin hollow nerve conduits. Moreover, we critically evaluated the cytotoxicity and differentiation of encapsulated neuroblastoma cells (N2a) toward neurons in culture with differentiation inducers.
Taken together, our study unequivocally establishes a significant step forward in developing a broad spectrum of shape-fidelity compliant, biocompatible bioink for the 3D bioprinting of biomimetic bone, cartilage, and neural scaffolds
Integrated Optic Waveguide Bragg Grating Devices in Silicon-on-Insulator and Their Applications: Design, Analysis and Fabrication
No full textIntegrated Waveguide Bragg Gratings are on-chip counterparts of Fiber Bragg Gratings. The working principles are similar, and functionalities could be explained with Coupled Mode Theory. Periodic index contrast required for Bragg reflection is achieved through physical corrugation of the waveguide walls. Being an 1D photonic crystal, such an Integrated Optic component is necessary for numerous applications involving wavelength selective functionality.
Due to the large index contrast and tight mode confinement of sub-micron waveguides, realization of narrow bandwidth Bragg Gratings on chip remains a difficult task. In this thesis, we study the integration of Bragg gratings on different Silicon Photonic waveguide configurations and engineer the structures to design devices for communication and sensing applications. The work covers intuitive conceptualizations, novel designs, and also materials and methods of fabrication and characterization of few devices. Firstly, Bragg Gratings on conventional Photonic Wire Waveguides are designed and analyzed. Further, shallowly etched ribs, deeply etched ribs (DER) and slot waveguides are optimized for the incorporation of sidewall corrugations.
Wavelength filters with narrow bandwidth, smaller device footprint, lesser complexity and higher fabrication tolerance are essential for on-chip Wavelength Division Multiplexing (WDM) applications. As part of this work, we design novel sub-micron, DER waveguide to achieve ultra narrow bandwidth gratings. The perturbations applied on thin slab sidewall eliminates the requirement of narrow corrugation width, which is otherwise possible only with high resolution lithography systems. 3-dB bandwidths as narrow as 0.5 nm at 1550 nm, desirable for DWDM are achieved. Further we analyze DER grating designed on Silicon Nitride platform and model its surface adsorption sensing capability for electrostatic assembly of Poly Electrolyte Multilayers (PEM). Following this, we focus on novel Silicon on Insulator (SOI) slot waveguide geometries. A dual slot waveguide is optimized in terms of induced geometrical asymmetry, to tailor the dispersion parameters. We observe that Bragg gratings on vertical slot waveguides exhibit a stop-band closure behavior. This phenomenon is utilized to bring down the transmission bandwidth of slot gratings, which is then used to design a refractive index sensor. In comparison to conventional vertical slots, loaded slots are found to have better confinement factor and are optimized to achieve polarization independent behavior. Further we implement Bragg gratings on sidewalls of loaded slots to obtain narrow bandwidths
Accurate Prediction of Enhancement Factors for Water Flow Through Boron Nitride Nanotubes
No full textWater in nanoconfined spaces, such as nanotubes, exhibit anomalous yet intriguing behaviour compared to bulk water, a better understanding of which can enable us to realize a sustainable future. Nanotubes are atomically thin sheets (e.g., graphene or hexagonal boron nitride) that have been rolled into tubes. Boron nitride nanotubes (BNNTs) have been explored for a wide variety of applications ranging from water desalination to osmotic power harvesting since their prediction and experimental discovery in 1994 and 1995, respectively. However, even after three decades of research, water flow through BNNTs is not fully understood at a fundamental level. In this thesis, we considered several aspects that were not given enough attention in previous studies of nanoconfined flow through BNNTs. For instance, no simulation work has modelled the changes in the partial charge distribution when a flat sheet is rolled into a tube, up to this point. To address this knowledge gap, we employed electronic density functional theory (DFT) calculations to accurately estimate quantum-mechanically derived partial charges on boron (B) and nitrogen (N) atoms in BNNTs of varying lengths and diameters. We observed a spatially varying charge distribution inside both armchair and zigzag nanotubes of finite length. Performing DFT calculations for longer BNNTs is computationally intractable even using state of the art resources. To solve this issue, we performed DFT calculations for shorter BNNTs and devised a charge assignment scheme to predict partial charges for longer BNNTs, thus overcoming the need to perform expensive DFT calculations. Subsequently, we performed molecular dynamics (MD) simulations to calculate enhancement factors (EFs), that quantify the extent to which the Hagen-Poiseuille equation is disobeyed at the nanoscale, for BNNTs of varying lengths and diameters. To elucidate the effects of electrostatic interactions, we used three different kinds of partial charge distributions on B and N atoms in a BNNT: (i) bulk partial charges from pristine hBN sheets (±0.907e, where e is the magnitude of charge on an electron), (ii) accurate partial charges obtained from DFT calculations, and (iii) the typical partial charge on carbon atoms in carbon nanotubes (0.0e). BNNTs with the bulk and zero partial charges exhibited the lowest and the highest flow enhancements, respectively, whereas those with accurate partial charges had intermediate EFs. We also incorporated atomic vibrations into our study and discovered, surprisingly, that these vibrations lead to a reduction in the water flow through BNNTs. Finally, we also investigated the effect of vacancy defects in a BNNT on water flow and observed that a single boron and diboron vacancy defects do not affect water flow if atomic vibrations are considered. Our results demonstrate the combined role of atomic vibrations, electrostatic interactions, and defects in modulating water flow through BNNTs and unravel partially the reasons for ultra-low flow EFs in BNNTs. Overall, we believe that the insights developed in this thesis can aid in the fabrication of tailor-made nanofluidic devices which can be employed for sustainability applications in the upcoming decades
Investigations on deep-level defects in HgTe nanocrystals-based photovoltaic devices using a novel instrumentation for Deep Level Transient Spectroscopy
No full textColloidally produced nanocrystals (NCs) arranged in thin films hold promise for next-generation semiconductors. These NCs offer tunability in semiconductor properties due to their size, shape, composition, and surface characteristics. However, the performance of NC-based optoelectronic devices still lags behind theoretical predictions. This is primarily attributed to electronic deep-level trap states, which act as recombination centres and limit effective mobility. The large surface area, hybrid nature, and disordered structure of NCs contribute to the abundance of trap states. To improve device performance, it is crucial to identify these defects and understand their impact on electrical characteristics. This work employs Deep Level Transient Spectroscopy (DLTS) to identify deep-level defects in NCs and NC-based photovoltaic devices. DLTS allows for determining defect level energy, concentration, capture cross-section, and differentiation between minority and majority carrier traps. This technique is highly sensitive, capable of detecting low defect concentrations, and resolves signals from various traps.
The conventional DLTS system suffers from drawbacks, including the need for multiple temperature cycles, which can lead to poor device contact and thin film adhesion. Additionally, maintaining a consistent temperature environment for each measurement is challenging, resulting in low-quality data. To address these issues, we develop a microcontroller-based DLTS system. This system utilizes a capacitance meter and electronic circuits controlled by an Arduino-Due microcontroller. We have used Arduino-Due to generate the filling pulse, monitor the capacitance, temperature, data acquisition, timing control and signal processing. By conducting measurements within a single temperature scan, our system saves time, improves accuracy, and reduces experimental failures. We validate the innovative instrumentation using a gold-doped silicon p-n junction sample. Furthermore, we apply this microcontroller-based DLTS system to study deep-level defects in Mercury Telluride (HgTe) nanocrystal-based photovoltaic devices.
We fabricate photovoltaic devices based on HgTe NCs/TiO2 and employ capacitance-voltage (C-V) and DLTS techniques to investigate and collect quantitative data on deep-level trap states. DLTS confirms the presence of interface trap states, while frequency-dependent capacitance measurements support the influence of charge storage in these nanocrystal-based heterostructures, offering insights for advanced device development. Using DLTS, we measure trap energy, capture cross-section, and concentration. These traps in the photovoltaic devices can act as recombination centres and effectively interact with valence and conduction bands. Poor device responsiveness is observed in the ITO/TiO2/HgTe/Au configuration due to inefficient photo charge extraction.
To enhance device performance, we optimize hole and electron extractions by introducing a Molybdenum Oxide (MoO3) hole extraction layer. We investigate the effect of this contact layer on trap level formation in the FTO/TiO2/HgTe/MoO3/Au photovoltaic device using low-temperature I-V, C-V, C-F, and microcontroller-based DLTS measurements. The obtained trap energy levels are comparable to those of the ITO/TiO2/HgTe/Au device, indicating the presence of trap levels at the TiO2/HgTe interface and no significant impact of the MoO3 contact layer on trap formation. Our microcontroller-based DLTS system proves to be an efficient tool for determining defect levels in heterojunctions based on nanocrystals. Surface states at the HgTe nanocrystals and oxygen vacancies in TiO2 are identified as the main contributors to trap levels, primarily located at the TiO2/HgTe interface. To further confirm the origin of trap states, we fabricate an ITO/HgTe/Al Schottky junction and measure the defect level energy using low-temperature I-V and C-F measurements. The obtained energy values support trap levels resulting from surface reconstruction at the TiO2/HgTe heterojunction interface. Passivating these trap states is crucial for improving device effectiveness
Interaction of Ionising Radiations with Nanoparticles
Get PDFThe interaction of ionizing radiations such as alpha, beta, gamma, and X-rays with matter-at bulk has been studied intensively for many decades. However, the interaction of ionizing
radiations with matter-at-nanoscale is studied sparsely due to the lack of experimental
techniques. Thus, there exists a gap in knowledge. The present thesis contributes to the
development of an experimental technique for determining the outcomes of the interaction of
given ionizing radiation with given nanoparticles. The technique involves obtaining pulse
height spectra of ionizing radiation with a liquid scintillator before and after loading the
nanoparticles under identical conditions and observing the variations in spectra to infer the
outcomes of the interactions.
The study investigates the outcomes of interactions of gamma-rays, X-rays, beta- and alpha radiations with about twenty-five types of nanoparticles. It ascertains the effects of the nature
and energy of radiations, species, size, and concentration of nanoparticles on the outcome of
interactions. It demonstrates that the interaction of ionizing radiations with nanomaterials
differs from those with their bulk counterparts. The interaction of low-energy photons (X-rays
from 55Fe or a 40 kVp gun or gamma-rays from 241Am or 133Ba) with nanoparticles of Gd2O3,
HfO2, and ZrO2 leads to the emission of numerous electrons from the nanoparticles. However,
the nanoparticles of Au, Fe2O3, Pd, W, and WO3 interact with low-energy photons but inhibit
the exit of electrons from them. Thus, the interaction of low-energy photons varies with the
species of nanoparticles. Further, photons of a given energy range interact with the
nanoparticles intensely. These are the two new results from this study. High-energy gamma
radiations seldom interact with nanoparticles. The interactions of beta- and alpha-radiations
result in the emission of electrons from all species of nanoparticles.
Practical applications like –nanoparticle radiosensitization for cancer treatment; the
development of efficient-fast-large-affordable gamma-detectors; and the development of Pb free, efficient, light-weight gamma-ray shields—rely on the interaction of ionizing radiations
with nanoparticles. They either seek or benefit from empirical knowledge of the outcome of
interactions. As the lack of mechanistic understanding of nanoparticle radiosensitization has
delayed its field implementation, researchers seek the outcomes of ‘physical interaction of
ionizing radiations with nanomaterials’. Since the process-related challenges have hindered the
upscaling of detectors or shields and have kept their studies in exploratory mode, certainty
gained on the outcome of interactions offers much-needed directions.
Nanoparticles of Gd2O3, HfO2, and LaF3 suit as dopants in plastic scintillators for developing
efficient-fast-large-affordable gamma detectors. Those of WO3, Sn, and Fe2O3 suit as dopants
for developing Pb-free, efficient gamma-ray shields. The results reason why the enhancement
of photon detection efficiency of plastic scintillators is repeatedly reported with doping of only
selected species of nanoparticles. They reason how nanoparticle-loaded polymers offer
impressive shielding efficiencies for diagnostic photons
Structural Health Monitoring Accounting for Thermal Variability and Damage using Approximate Bayesian Computation
No full textIn structural engineering, damage is characterized as a change in material property,
boundary condition, or geometry. The changes in these properties/parameters lead to
a change in the measured response. The difference in measurements can be due to
actual damage in the member (due to crack formation, corrosion of rebars, or crushing
of concrete), or it might be due to temperature variations while making measurements.
Temperature variability significantly affects the accuracy of structural health monitoring
strategies in quantifying structural damage. Performing damage detection without
isolating/incorporating these variations can lead to false damage detection, i.e., the
undamaged structure can be detected as damaged. Hence, a method is required to isolate
the effect of these variabilities while detecting damage. Researchers have developed
methods to analyze and separate the effects of environmental variability from damageinduced changes in the measures. The main two approaches are (a) data-based, which
uses statistics-based tools for analyzing patterns in the data or compute parameters,
and (b) model-based, where the method considers both environmental and damagebased changes of stiffness value. This study uses a model-based approach to address
the problem of detecting damage under different temperature levels in undamaged and
damaged states. The proposed method uses an Approximate Bayesian computation
Nested Sampling (ABC-NS) algorithm to detect damage under temperature variability.
The study introduces a new damage index for identifying potentially damaged members.
After performing damage localization, we estimate the parameters’ posterior distribution
for potentially damaged members using ABC-NS. The estimated parameters’ mean value
corresponds to the parameters’ actual values in the damaged state. In this study, we
will see how to incorporate the effect of temperature variation and noise using a finite
element model. One of the major assumptions in a lot of studies is that the structure
remains in an equivalently linear regime accounting for damage. However, a breathing
crack can lead to bi-linear stiffness and affect structural health monitoring strategies,
classified as damage-induced nonlinearity. This study also incorporates damage-induced
nonlinearity while performing damage detection
Learning to Perceive Humans From Appearance and Pose
No full textAnalyzing humans and their activities takes a central role in computer vision. This requires machine learning models to encapsulate both the diverse poses and appearances exhibited by humans. Estimating the 3D poses of highly deformable humans from monocular RGB images remains an important, challenging, and unsolved problem with applications in human-robot interaction, augmented reality, gaming industry, etc. Another important task is to identify the same human targets across camera viewpoints in a wide-area video surveillance setup, requiring learning discriminative and robust representations of human appearances under large variabilities of poses, backgrounds, and illuminations. In this thesis, we study several computer vision problems under the theme of estimating human pose and modeling appearance from monocular images.
Estimating the 3D pose from a single image is an ill-posed classical inverse problem as the model lacks depth information. In such scenarios, supervised approaches tend to perform well by guiding the model towards plausible poses. While assuming the availability of a labeled dataset is itself impractical, such approaches tend to suffer from poor generalization to unseen datasets. We thus formulate the problem as an unsupervised learning task and propose a novel framework that consists of a series of differentiable transformations acting as a suitable bottleneck, stimulating effective pose disentanglement. Furthermore, the proposed adaptation technique enables learning from in-the-wild videos beyond laboratory settings, thereby resulting in superior generalizability across diverse and unseen environments.
The 3D pose estimation models discard variations in a human body, e.g., shape and appearance, which may help solve other related tasks such as body-part segmentation. As a next step, we design a single part-based 2D puppet model, relying on human pose articulation constraints and a set of unpaired 3D poses to estimate both 3D poses and part segments from human-centric images. Unlike our previous work, the proposed part-based model allows us to operate on videos with diverse camera movements.
The approaches above cast the 3D pose estimation problem as a task of disentangling human pose and appearance. Different from these, we propose to cast the 3D pose learning as a cross-modal alignment problem in our subsequent work. We consider the availability of an unpaired pool of short-length natural action videos and 3D pose sequences from the input and output modalities respectively. We introduce a novel technique for self-supervised alignment across these modalities while relying on preserving higher-order non-local relations in a pre-learned, latent pose space to attain superior generalizability over the state-of-the-art.
Unsupervised person re-identification (re-ID) aims to tackle the problem of matching identities across non-overlapping cameras without any assumption of labels during training. We propose a two-stage training strategy towards solving this task. First, we train a deep network on an expertly designed pose-transformed dataset obtained by generating multiple perturbations in the pose space for each original image. Next, the network learns to attend to the fundamental aspects of feature learning - compact clusters with low intra-cluster and high inter-cluster variation, thereby mapping similar features closer using the proposed discriminative clustering algorithm. Experiments on large-scale re-ID datasets demonstrate the superiority of our method against state-of-the-art approaches