121 |
Scalable Register File Architecture for CGRA AcceleratorsJanuary 2016 (has links)
abstract: Coarse-grained Reconfigurable Arrays (CGRAs) are promising accelerators capable
of accelerating even non-parallel loops and loops with low trip-counts. One challenge
in compiling for CGRAs is to manage both recurring and nonrecurring variables in
the register file (RF) of the CGRA. Although prior works have managed recurring
variables via rotating RF, they access the nonrecurring variables through either a
global RF or from a constant memory. The former does not scale well, and the latter
degrades the mapping quality. This work proposes a hardware-software codesign
approach in order to manage all the variables in a local nonrotating RF. Hardware
provides modulo addition based indexing mechanism to enable correct addressing
of recurring variables in a nonrotating RF. The compiler determines the number of
registers required for each recurring variable and configures the boundary between the
registers used for recurring and nonrecurring variables. The compiler also pre-loads
the read-only variables and constants into the local registers in the prologue of the
schedule. Synthesis and place-and-route results of the previous and the proposed RF
design show that proposed solution achieves 17% better cycle time. Experiments of
mapping several important and performance-critical loops collected from MiBench
show proposed approach improves performance (through better mapping) by 18%,
compared to using constant memory. / Dissertation/Thesis / Masters Thesis Computer Science 2016
|
122 |
Modélisation du dichroïsme circulaire des protéines : modèle simple et applications / Modelisation of protein circular dichroism : simple model and applicationTran, Viet-Dung 18 December 2015 (has links)
La spectroscopie de dichroïsme circulaire (CD) est une des techniques fondamentales en biologie structurale qui permet la détermination du contenu en structures secondaires d'une protéine. Le rayonnement synchrotron a considérablement augmenté l’utilité de la méthode, car il permet de travailler avec une gamme spectrale étendue et à meilleure intensité. Le développement de modèles permettant d’établir une relation entre la structure d’une protéine et son spectre CD d’une manière efficace n’a pourtant pas suivi l’évolution technique et l’analyse de spectres CD de protéines entières reste un défi sur le plan théorique. Dans ce contexte, nous avons développé un modèle "minimaliste" pour la spectroscopie CD des protéines, où chaque atome C-alpha de la chaîne principale porte un oscillateur de Lorentz classique, i.e. une charge mobile qui est tenue par un potentiel quadratique. Les oscillateurs sont couplés par un potentiel coulombien et leurs déplacements suivent les tangentes locales respectives de la courbe spatiale décrite par les atomes C-alpha. Le système d'oscillateurs est couplé à une onde électromagnétique plane décrivant la source de lumière et le phénomène d'absorption est modélisé par des forces de friction. Nous montrons que le modèle reproduit correctement le phénomène CD d'une chaîne polypeptidique hélicoïdale et en particulier son signe en fonction de l'orientation de la chaîne. Comme première application, nous présentons l'ajustement du modèle au spectre CD d'un polypeptide composé de 15 résidus qui se plie sous forme d'une hélice alpha. La transférabilité de ces paramètres est ensuite évaluée pour la myoglobine, une protéine de 153 résidus contenant 8 hélices alpha. / Circular dichroism (CD) spectroscopy is one of the fundamental techniques in structural biology that allows us to investigate the secondary structure of proteins. Synchrotron radiation has considerably increased the usefulness of the method because it allows to work with a wider range of spectrum and much greater signal-to-noise ratios. The development of a theoretical model to establish a relationship between the structure of a protein and its CD spectra in an efficient manner proved to be a complex task. The calculation of the CD spectra of large molecules, such as protein, remains a challenge, due to the size and flexibility of the molecules. In this context, we have developed a “minimal” model to explain the CD spectroscopy of proteins, which associates each C-alpha position on the protein backbone with a classical Lorentz oscillator i.e. a mobile charge attaches to a corresponding atom by a quadratic potential. The coupling between charges is through the Coulomb potential and their displacements follow the direction of the respective local tangents to the Calpha space curve. This system is coupled to a planar electromagnetic wave describing the light source and the absorption phenomenon is modeled by frictional forces. We show that the model correctly reproduces the CD phenomenon of a helical polypeptide chain and in particular its sign depending on the orientation of the chain. At first, we have fitted a model to CD spectra of a polypeptide chain of 15 residues folded into alpha helix. The transferability of these parameters is then evaluated with myoglobin, a protein of 153 residues containing eight alpha helices.
|
123 |
From coarse-grained to atomistic molecular modeling : how structure and dynamics shape intra-molecular communication and functional sites in proteins / Du gros-grains à la modélisation moléculaire tout-atome : comment la structure/dynamique façonnent la communication intramoléculaire et les sites fonctionnels dans les protéinesAubailly, Simon 27 January 2017 (has links)
Dans cette thèse, nous nous sommes intéressés à la relation subtile qui existe entre lesstructures complexes des protéines et leurs fonctions encore plus raffinées que ces dernièreseffectuent. Basés sur deux descriptions différentes des protéines, à l’échelle de acide-aminé età l’echelle atomique, un de nos objectifs était de connecter des indicateurs structuraux calculésà partir de la topologie des protéines à des sites fonctionnels tels que les sites catalyiquesdans les enzymes. Un autre pan de la recherche de cette thèse était d’utiliser nos outils baséssur la structure et de mettre au point de nouvelles simulations numériques pour étudier lesdéterminants basiques structuraux et dynamiques de la communication intramoléculaire dansles protéines. Une première découverte fut de montrer comment l’analyse des modes normauxet la théorie des reseaux complexes conduisent à la prédiction des sites catalytiques dans lesenzymes. De plus, nous avons travaillé sur un groupe relativement peu connu de modes nor-maux qui ont la particularité d’être localisés à deux endroits très eloignés dans la structure desprotéines. Ces modes bilocalisés ont permis de réaliser des transferts d’énergie à des distancesconsidérables (plus de 70 Å). Finalement, des expériences de refroidissement effectuées sur unsystème protéine-eau décrit à l’échelle atomique ont dévoilé que le refroidissement induit unelocalisation spontanée d’énergie, indiquant certaines déformations des anneaux du benzenecomme possible centres de stockage de l’énergie dans les protéines. / In this thesis we have focused on the elusive relation that exists in proteins between theircomplex structures and the even more complex and sophisticated functions that they perform.Based on two different descriptions of proteins, at residue and atomistic scale, one of ouraims was to connect structural indicators computed from the topology of protein scaffoldsto hot spots in proteins such as catalytic sites in enzymes. Another goal of this thesis wasto employ our structure-based tools and set up original simulation scheme to investigate thebasic structural and dynamical determinants of intramolecular communication in proteins.As a first important finding, we have shown how normal mode analysis and specific graph-theoretical approaches lead to the prediction of catalytic sites in enzymes. Moreover, wehave concentrated our attention on an overlooked class of normal modes, that are stronglylocalized at two widely separated locations in protein scaffolds. These bilocalized modesturned out to efficiently mediate energy transfer even across considerable distances (morethan 70 Å). Finally, cooling experiments performed on a protein-water system described atatomic level have unveiled complex cooling-induced spontaneous energy localization patterns,pointing to specific deformation modes of benzene rings as potential energy-storage centers.
|
124 |
Formation des chondres et relation avec leurs auréoles de matrice à grains fins / Chondrule formation and relation with their fine-grained rimsSoulié, Camille 10 December 2014 (has links)
Les météorites les plus primitives, appelées chondrites, sont les témoins privilégiés de la formation des premiers solides du système solaire et des processus qui ont conduit à la naissance des premiers corps planétaires. Les chondres, qui sont des sphérules ferromagnésiennes ignées, et leur matrice finement cristallisée constituent l’essentiel du volume des chondrites primitives. L’origine des chondres comme celle de la matrice est encore largement débattue. Cette thèse interroge les relations qui existent entre les chondres et la matrice à grains fins qui les auréole fréquemment, notamment dans les chondrites carbonées. En prenant l’exemple de la chondrite Vigarano (CV3) nous explorons la piste d’un continuum des conditions physico-chimiques pour former les chondres et leurs auréoles de matrice à grain fin. La thèse aborde cette problématique par une double approche : une approche expérimentale consacrée à la quantification des vitesses de dissolution de l’olivine dans des liquides silicatés caractéristiques des chondres de type I et une approche de caractérisation détaillée de la minéralogie, des fabriques et de la composition isotopique de l’oxygène des chondres et de leurs auréoles de matrice à grains fins dans Vigarano. Ces travaux suggèrent que chondre et matrice pourraient enregistrer une continuité d’interaction avec le gaz environnant, d’abord de type liquide/gaz (solubilité) pour la formation des chondres, puis de type solide/gaz (condensation) à plus basse température pour la formation des auréoles de matrice à grains fins / The most primitive meteorites, the chondrites, are our best witnesses of solar system birth and formation of first solids as well as processes that formed the first planetary bodies. Both chondrules, which are ferromagnesian igneous spherules, and fine-grained matrix, constitute the largest volume fraction of primitive chondrites. Their origin still constitutes a matter of debate. This thesis questions the relationship between chondrules and fine-grained matrix, called fine-grained rims, that frequently overlay chondrules in primitive carbonaceous chondrites. By taking the example of Vigarano (CV3), we explore the idea that chondrules and fine-grained rims could form in a continuum of physical and chemical conditions. This thesis combines both experimental and characterization approaches, with an experimental determination of olivine dissolution rates in molten silicates representative of type I chondrule melts and a detailed petrographic characterization of mineralogy, fabrics and oxygen isotopic composition of chondrules and their fine-grained rims in Vigarano. This work suggests that chondrules and their fine-grained rims could register a continuous interaction with surrounding gas, with first, chondrule formation from liquid/gas interaction (solubility) and then fine-grained rim formation, at lower temperature and from solid/gas interaction (condensation)
|
125 |
Improved understanding of sublevel blasting : Determination of the extent of the compacted zone, its properties and the effects on cavingPetropoulos, Nikolaos January 2017 (has links)
Sublevel caving (SLC) is a mass mining method relying on the flowability of the blasted material. The ore is blasted in slices against caved material which is mainly waste rock. The result of the confined blast is greatly influenced by the interaction between the blasted material and the caved material. During blasting both materials change characteristics; the blasted material increases its porosity and compressibility due to breakage and swelling while the caved material is compacted and decreases in porosity and compressibility. The understanding of the mechanisms involved in this process is of significant importance. The behavior of the caved material (confining material) was studied in laboratory under dynamic loading. A new apparatus was developed to conduct impact tests to simulate blasting conditions. The tested material was a blend of crushed waste rock from drift development in the Kiirunavaara mine with maximum particle size 32 mm. The material was tested for two conditions, i.e. dry and wet (pendular state), and with different impact velocities (low (5 m/s), medium (8 m/s) and high (10-12 m/s)). During the impact tests, two types of measurements were taken; dynamic measurements based on the recordings from the installed accelerometers on the machine and static measurements pre- and post-impact. Additionally, the angle of repose, the impact duration, and the fragmentation was measured. In addition to the laboratory tests, small-scale blasting tests were carried out to investigate the burden behavior in confined conditions. The blasted specimen was a cuboid magnetic mortar block and the confining material was crushed concrete with maximum particle size 16 mm. The blocks were instrumented with custom-made incremental displacement sensor. After the analysis of the results from the above experimental work, two confined pillar tests (test #1 and test #2) were carried out at the Kiirunavaara mine. The preparation work for the pillar tests involved the development of instrumentation and installation techniques. The experimental configuration contained two blastholes and measurement holes in between the blastholes drilled from the neighboring drift. Test #1 mainly focused on the evaluation of the instrumentation and techniques while test #2 was focused on the interaction between the blasted burden and the confining material. The confining material in test #1 was a blend of ore and waste material from drift development at the Kiirunavaara mine. The characteristics of the material were unknown. Test #2 was split into two parts, the confining material in the first part was the same as in the laboratory impact tests and the second part of the pillar was confined by caved masses. The instrumentation was installed in the burden of the pillars and was equipped with accelerometers and displacement sensor. Additional instrumentation was also installed in the confining material. The burden in the small-scale blasting tests reached maximum velocity 29 m/s and maximum displacement 12.6 mm. In pillar tests, the burden movement was in the range of 0.9 to 1.1 m. In both pillar tests, burden erosion material was observed in the gap between the intact and the blasted burden. This material was finer compared to the blasted burden. The origin of this material was from the vicinity of the blastholes. The results of the laboratory tests showed that the wet material exhibited larger compaction zone than that of the dry material. The wet material showed apparent cohesion close to the impact surface of the tested material. A similar observation was made in test #2 where an agglomeration of the confining material, as a result of apparent cohesion, was observed on the surface of the blasted burden. The displacement data from the instrumentation in the burden and inside the confining material showed that the compaction zone follows an inverse exponential behavior. After the blast steeper angles of repose were measured indicating higher frictional forces between the particles. Moreover, the evidence of apparent cohesion and a larger angle of repose indicated the introduction of tensile strength in the material. The mass of the confining material was compressed elastically and plastically during the blast. After the blast, the material recovered its elastic deformation and pushed the blasted burden backward as observed in the small-scale blasting tests and the pillar tests. At this stage, the burden erosion material was compacted. Hence, there were 3 materials, i.e. burden erosion material, burden and confining material, which were compacted with different compaction rates. This condition promotes interlocking of the particles in the materials. If this behavior is correlated with a production SLC ring, then it indicates disturbances in flowability of the blasted material.
|
126 |
Um controle de versões refinado e flexível para artefatos de software / Flexible and fine-grained version control for software artifactsDaniel Carnio Junqueira 07 January 2008 (has links)
As atividades de controle de versões são consideradas essenciais para a manutenção de sistemas de computador. Elas começaram a ser realizadas na década de 1950 de forma manual. As primeiras ferramentas de controle de versões, que surgiram nos anos setenta, não evoluíram significativamente desde sua criação e, até hoje, o controle de versões de arquivos é geralmente realizado em arquivos ou mesmo módulos completos, utilizando os conceitos que foram lançados há mais de três décadas. Com a popularização da utilização de sistemas computacionais, ocorreu um sensível aumento no número de sistemas existentes e, também, na complexidade dos mesmos. Além disso, muitas alterações ocorreram nos ambientes de desenvolvimento de software, e existe demanda por sistemas que permitam aos desenvolvedores ter cada vez mais controle automatizado sobre o que está sendo desenvolvido. Para isso, algumas abordagens de controle de versões refinados para artefatos de software foram propostas, mas, muitas vezes, não oferecem a exibilidade de utilização exigida pelos ambientes de desenvolvimento de software. Neste trabalho, é apresentado um sistema que visa a fornecer suporte ao controle de versões refinado e flexível para artefatos de software, tendo por base um modelo bem definido para representação das informações da estrutura dos arquivos que compõem determinado projeto de software, sejam eles código-fonte dos programas de computador, documentação criada em Latex, arquivos XML, entre outros. O sistema apresentado foi projetado para ser integrado com outras soluções utilizadas em ambientes de desenvolvimento de software / Version control tasks are considered essential for the maintenance of computers systems. They have been done since beginning of 50\'s in a by hand manner. First tools, which were released in 70\'s, didn\'t evolve significantly since its creation, and, in general, version control systems still work with entire files or even modules of software, having the same concepts that were launched more than three decades ago. With the popularization of computers systems there had been a sensible increase in the number of existing systems and also in the complexity of these systems. Besides that many changes have taken place in the software development environments, and there is demand for systems which allow developers to have more automated control about what is being developed. Regard to this demand some approaches of fine-grained version control have been proposed, but they usually do not provide the required exibility for its use in the real software development environments. In this work its presented a system which aims at providing support for exible and fine-grained version control of software artifacts, using a well defined model to represent the logical structure of the files which compose a software project, independently of its type - they can be XML files, source-code files, Latex files and others. The system has been designed to be integrated with other software solutions used in software development environments
|
127 |
Processos, fácies e geometria do sistema turbidítico da formação Taciba/Membro Rio Segredo, faixa aflorante norte catarinense /Andrade, Lygia Rodrigues de Moraes de. January 2009 (has links)
Orientador: Joel Carneiro de Castro / Banca: Maria Rita Caetano Chang / Banca: Luiz Carlos Weinschütz / Resumo: Os membros Lontras (Formação Campo Mourão) e Rio Segredo (Formação Taciba) constituem uma sucessão marinha de folhelhos e arenitos turbidíticos, que está encaixada entre diamictitos glaciais daquelas formações. O folhelho Lontras tem uma centena de metros na faixa aflorante norte catarinense, contendo em sua porção superior a fácies "folhelho várvico", na verdade um estrato com gradação de siltito a folhelho (Tde) em escala milimétrica e que é considerado como sendo o turbidito distal da sucessão. O turbidito Rio Segredo tem de 15 a 25 m e consiste de estratos gradacionais portando sequência Bouma em diversas escalas: muito delgado (1 a 3 cm; Tde e Tcde), delgado (3 a 10 cm; Tcde e Tbcde), médio (10 a 30 cm; Tbcde e Tabcde), espesso (30 a 100 cm; Tabc) e muito espesso (acima de 1,0 m; Tabc). Há ainda uma divisão "superior" do Membro Rio Segredo normalmente com 10 a 20 m de espessura, que contém turbiditos areno-argilosos muito delgados (Tcde e Tde). Foram levantados sete perfis faciológicos de detalhe, escala 1:50, de modo a registrar turbiditos com até 5 cm de espessura (1 mm no perfil). Posteriormente, quatro desses perfis foram cronocorrelacionados em uma seção norte-sul: Forcação, Wiegand, Laeisz e Dona Emma, numa extensão de 28 km. A esta seção foi acrescido o perfil Taiózinho, localizado 30 km a oeste do Forcação, entre os dois últimos (semelhança faciológica com o Laeisz). Identificaram-se sete sistemas deposicionais, com uma média de 3 m de espessura por sistema e contidos em três sequências de alta frequência. Os sistemas deposicionais são formados por ciclos turbidíticos de origem marinha, encontrados principalmente nos perfis Laeisz e Taiózinho, com uma organização ascendente de adelgaçamento e granodecrescência. Outros ciclos turbidíticos... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The Lontras shale and overlying Rio Segredo sandstone are members of Campo Mourão and Taciba formations, corresponding to shelf marine and turbidite deposits; they are encased in glacial diamictites of those formations. The Lontras shale is 100 m thick in northern Santa Catarina outcrop belt and it displays a "varved shale" facies in its upper portion. In reality, it represents many mm-scale beds with Bouma sequence Tde, and therefore distal, argillaceous turbidites. The Rio Segredo Member is 15 to 25 m thick and contain beds of different thicknesses: very thin (1 to 3 cm; Tde and Tcde), thin (3 to 10 cm; Tcde and Tbcde), and medium beds (10 to 30 cm; Tbcde and Tabcde). Also, thick (30 to 100 cm) and very thick beds (thicker than 1 m) displays massive or graded sandstones with disperse laminations and cross-laminations resembling Ta, Tb and Tc intervals of Bouma sequence. There is also a Rio Segredo "upper" division with 10 to 20 m thick, consisting of very thin turbidites (Tcde and Tde). Seven detailed facies logs were constructed at 1:50 scale, to represent even 5 cm thin beds (1 mm). Later, four of the logs were put in a north-south stratigraphic section (28 km in length): Forcação, Wiegand, Laeisz and Dona Emma, in a 28 km extension. A fifth log, Taiózinho (distant 30 km west of Forcação), was added to the section between the last two logs, because of its facies similarity with Laeisz log. Seven depositional systems are identified, averaging 3 m thick in, thickness, and they belong to high frequency sequences. The systems are composed of turbidite cycles of marine origin, found mainly in Laeisz and Taiózinho logs: they form thinning- and fining-upward cycles. Turbidite cycles of deltaic origin are observed in Forcação and Dona Emma logs as thickening-up and coarsening-up cycles. In the latter, are included... (Complete abstract click electronic access below) / Mestre
|
128 |
Une méthode hybride pour la classification d'images à grain fin / An hybrid method for fine-grained content based image retrievalPighetti, Romaric 28 November 2016 (has links)
La quantité d'images disponible sur Internet ne fait que croître, engendrant un besoin d'algorithmes permettant de fouiller ces images et retrouver de l'information. Les systèmes de recherche d'images par le contenu ont été développées dans ce but. Mais les bases de données grandissant, de nouveaux défis sont apparus. Dans cette thèse, la classification à grain fin est étudiée en particulier. Elle consiste à séparer des images qui sont relativement semblables visuellement mais représentent différents concepts, et à regrouper des images qui sont différentes visuellement mais représentent le même concept. Il est montré dans un premier temps que les techniques classiques de recherche d'images par le contenu rencontrent des difficultés à effectuer cette tâche. Même les techniques utilisant les machines à vecteur de support (SVM), qui sont très performants pour la classification, n'y parviennent pas complètement. Ces techniques n'explorent souvent pas assez l'espace de recherche pour résoudre ce problème. D'autres méthodes, comme les algorithmes évolutionnaires sont également étudiées pour leur capacité à identifier des zones intéressantes de l'espace de recherche en un temps raisonnable. Toutefois, leurs performances restent encore limitées. Par conséquent, l'apport de la thèse consiste à proposer un système hybride combinant un algorithme évolutionnaire et un SVM a finalement été développé. L'algorithme évolutionnaire est utilisé pour construire itérativement un ensemble d'apprentissage pour le SVM. Ce système est évalué avec succès sur la base de données Caltech-256 contenant envieront 30000 images réparties en 256 catégories / Given the ever growing amount of visual content available on the Internet, the need for systems able to search through this content has grown. Content based image retrieval systems have been developed to address this need. But with the growing size of the databases, new challenges arise. In this thesis, the fine grained classification problem is studied in particular. It is first shown that existing techniques, and in particular the support vector machines which are one of the best image classification technique, have some difficulties in solving this problem. They often lack of exploration in their process. Then, evolutionary algorithms are considered to solve the problem, for their balance between exploration and exploitation. But their performances are not good enough either. Finally, an hybrid system combining an evolutionary algorithm and a support vector machine is proposed. This system uses the evolutionary algorithm to iteratively feed the support vector machine with training samples. The experiments conducted on Caltech-256, a state of the art database containing around 30000 images, show very encouraging results
|
129 |
Computational Methods for the structural and dynamical understanding of GPCR-RAMP interactionsBahena, Silvia January 2020 (has links)
Protein-protein interaction dominates all major biology processes in living cells. Recent studies suggestthat the surface expression and activity of G protein-coupled receptors (GPCRs), which are the largestfamily of receptors in human cells, can be modulated by receptor activity–modifying proteins (RAMPs). Computational tools are essential to complement experimental approaches for the understanding ofmolecular activity of living cells and molecular dynamics simulations are well suited to providemolecular details of proteins function and structure. The classical atom-level molecular modeling ofbiological systems is limited to small systems and short time scales. Therefore, its application iscomplicated for systems such as protein-protein interaction in cell-surface membrane. For this reason, coarse-grained (CG) models have become widely used and they represent an importantstep in the study of large biomolecular systems. CG models are computationally more effective becausethey simplify the complexity of the protein structure allowing simulations to have longer timescales. The aim of this degree project was to determine if the applications of coarse-grained molecularsimulations were suitable for the understanding of the dynamics and structural basis of the GPCRRAMP interactions in a membrane environment. Results indicate that the study of protein-proteininteractions using CG needs further improvement with a more accurate parameterization that will allowthe study of complex systems.
|
130 |
Interpretable Fine-Grained Visual CategorizationGuo, Pei 16 June 2021 (has links)
Not all categories are created equal in object recognition. Fine-grained visual categorization (FGVC) is a branch of visual object recognition that aims to distinguish subordinate categories within a basic-level category. Examples include classifying an image of a bird into specific species like "Western Gull" or "California Gull". Such subordinate categories exhibit characteristics like small inter-category variation and large intra-class variation, making distinguishing them extremely difficult. To address such challenges, an algorithm should be able to focus on object parts and be invariant to object pose. Like many other computer vision tasks, FGVC has witnessed phenomenal advancement following the resurgence of deep neural networks. However, the proposed deep models are usually treated as black boxes. Network interpretation and understanding aims to unveil the features learned by neural networks and explain the reason behind network decisions. It is not only a necessary component for building trust between humans and algorithms, but also an essential step towards continuous improvement in this field. This dissertation is a collection of papers that contribute to FGVC and neural network interpretation and understanding. Our first contribution is an algorithm named Pose and Appearance Integration for Recognizing Subcategories (PAIRS) which performs pose estimation and generates a unified object representation as the concatenation of pose-aligned region features. As the second contribution, we propose the task of semantic network interpretation. For filter interpretation, we represent the concepts a filter detects using an attribute probability density function. We propose the task of semantic attribution using textual summarization that generates an explanatory sentence consisting of the most important visual attributes for decision-making, as found by a general Bayesian inference algorithm. Pooling has been a key component in convolutional neural networks and is of special interest in FGVC. Our third contribution is an empirical and experimental study towards a thorough yet intuitive understanding and extensive benchmark of popular pooling approaches. Our fourth contribution is a novel LMPNet for weakly-supervised keypoint discovery. A novel leaky max pooling layer is proposed to explicitly encourages sparse feature maps to be learned. A learnable clustering layer is proposed to group the keypoint proposals into final keypoint predictions. 2020 marks the 10th year since the beginning of fine-grained visual categorization. It is of great importance to summarize the representative works in this domain. Our last contribution is a comprehensive survey of FGVC containing nearly 200 relevant papers that cover 7 common themes.
|
Page generated in 0.066 seconds