Boosting-based transfer learning for multi-view head-pose classification from surveillance videos

This work proposes a boosting-based transfer learning approach for head-pose classification from multiple, low-resolution views. Head-pose classification performance is adversely affected when the source (training) and target (test) data arise from different distributions (due to change in face appearance, lighting, etc). Under such conditions, we employ Xferboost, a Logitboost-based transfer learning framework that integrates knowledge from a few labeled target samples with the source model to effectively minimize misclassifications on the target data. Experiments confirm that the Xferboost framework can improve classification performance by up to 6%, when knowledge is transferred between the CLEAR and FBK four-view headpose datasets.

Exploring Transfer Learning Approaches for Head Pose Classification from Multi-view Surveillance Images

Head pose classification from surveillance images acquired with distant, large field-of-view cameras is difficult as faces are captured at low-resolution and have a blurred appearance. Domain adaptation approaches are useful for transferring knowledge from the training (source) to the test (target) data when they have different attributes, minimizing target data labeling efforts in the process. This paper examines the use of transfer learning for efficient multi-view head pose classification with minimal target training data under three challenging situations: (i) where the range of head poses in the source and target images is different, (ii) where source images capture a stationary person while target images capture a moving person whose facial appearance varies under motion due to changing perspective, scale and (iii) a combination of (i) and (ii). On the whole, the presented methods represent novel transfer learning solutions employed in the context of multi-view head pose classification. We demonstrate that the proposed solutions considerably outperform the state-of-the-art through extensive experimental validation. Finally, the DPOSE dataset compiled for benchmarking head pose classification performance with moving persons, and to aid behavioral understanding applications is presented in this work.

An adaptation framework for head-pose classification in dynamic multi-view scenarios

Multi-view head-pose estimation in low-resolution, dynamic scenes is difficult due to blurred facial appearance and perspective changes as targets move around freely in the environment. Under these conditions, acquiring sufficient training examples to learn the dynamic relationship between position, face appearance and head-pose can be very expensive. Instead, a transfer learning approach is proposed in this work. Upon learning a weighted-distance function from many examples where the target position is fixed, we adapt these weights to the scenario where target positions are varying. The adaptation framework incorporates reliability of the different face regions for pose estimation under positional variation, by transforming the target appearance to a canonical appearance corresponding to a reference scene location. Experimental results confirm effectiveness of the proposed approach, which outperforms state-of-the-art by 9.5% under relevant conditions. To aid further research on this topic, we also make DPOSE- a dynamic, multi-view head-pose dataset with ground-truth publicly available with this paper.

ON THE UTILITY OF CANONICAL CORRELATION ANALYSIS FOR DOMAIN ADAPTATION IN MULTI-VIEW HEADPOSE ESTIMATION

The utility of canonical correlation analysis (CCA) for domain adaptation (DA) in the context of multi-view head pose estimation is examined in this work. We consider the three problems studied in 1], where different DA approaches are explored to transfer head pose-related knowledge from an extensively labeled source dataset to a sparsely labeled target set, whose attributes are vastly different from the source. CCA is found to benefit DA for all the three problems, and the use of a covariance profile-based diagonality score (DS) also improves classification performance with respect to a nearest neighbor (NN) classifier.

Identification of the nature of platinum related midgap state in silicon by deep level transient spectroscopy

In this article, we present the detailed investigations on platinum related midgap state corresponding to E-c -0.52 eV probed by deep level transient spectroscopy. By irradiating the platinum doped samples with high-energy (1.1 MeV) gamma rays, we observed that the concentration of the midgap state increases and follows a square dependence with irradiation dose. However, the concentration of the acceptor corresponding to E-c -20.28 eV remained constant. Furthermore, from the studies on passivation by atomic hydrogen and thermal reactivation, we noticed that the E-c -0.52 eV level reappears in the samples annealed at high temperatures after hydrogenation. The interaction of platinum with various defects and the qualitative arguments based on the law of mass action suggest that the platinum related midgap defect might possibly correspond to the interstitial platinum-divacancy complex (V-Pt-V).

A simple circuit to aid direct measurement of capture cross sections of deep level impurities

An inexpensive and simple circuit to aid the direct measurement of majority carrier capture cross sections of impurity levels in the band gap of a semiconductor by the variable width filling pulse technique is presented. With proper synchronisation, during the period of application of the pulse, the device is disconnected from the capacitance meter to avoid distortion of the pulse and is reconnected again to the meter to record the emission transient. Modes of operation include manual triggering for long emission transients, repetitive triggering for isothermal and DLTS measurements and the DLTS mode which is to be used with signal analysers that already provide a synchronising pulse for disconnection.

Deep inelastic electroproduction structure functions for polarized and unpolarized proton targets

Following Ioffe's method of QCD sum rules the structure functions F2(x) for deep inelastic ep and en scattering are calculated. Valence u-quark and d-quark distributions are obtained in the range 0.1 less, approximate x <0.4 and compared with data. In the case of polarized targets the structure function g1(x) and the asymmetry Image Full-size image are calculated. The latter is in satisfactory agreement in sign and magnitude with experiments for x in the range 0.1< x < 0.4.

New temperature fluctuation method for direct determination of thermal activation energy of deep levels in semiconductors

A new method is suggested where the thermal activation energy is measured directly and not as a slope of an Arrhenius plot. The sample temperature T is allowed to fluctuate about a temperature T0. The reverse-biased sample diode is repeatedly pulsed towards zero bias and the transient capacitance C1 at time t1 is measured The activation energy is obtained by monitoring the fluctuations in C1 and T. The method has been used to measure the activation energy of the gold acceptor level in silicon.

Stress Analysis of Deep Beams

In a beam whose depth is comparable to its span, the distribution of bending and shear stresses differs appreciably from those given by the ordinary flexural theory. In this paper, a general solution for the analysis of a rectangular, single-scan beam, under symmetrical loading is developed. The Multiple Fourier procedure is employed using four series by which it has been possible to satisfy all boundary and the resulting relation among the co-efficient are derives.

Electrical properties of nickel-related deep levels in silicon

Electrically active deep levels related to nickel in silicon are studied under different diffusion conditions, quenching modes, and annealing conditions. The main nickel-related level is at Ev+0.32 eV. Levels at Ev+0.15 and Ev+0.54 eV are not related to nickel while those at Ev+0.50 and Ev+0.28 eV may be nickel related. Their concentrations depend on the quenching mode. There is no nickel-related level in the upper half of the band gap. The complicated annealing behavior of the main nickel-related level is explained on the basis of the formation and dissociation of a nickel-vacany complex. Journal of Applied Physics is copyrighted by The American Institute of Physics.

Investigation of Deep Defects Due to -Particle Irradiation in n-Silicon

Electrical properties of deep defects induced in n-silicon by -particles of about 10 MeV energy at a dose of 1014 and 1015 cm-2 are studied by DLTS. The levels at Ec -0.18 eV, Ec -0.26 eV, and Ec -0.48 eV are identified as A center, V2 (=/-) and V2 (-/0) on the basis of activation energy, electron capture cross section, and annealing behavior. Two other irradiation related levels at Ec -0.28 eV and Ec -0.51 eV could not be related to any known center.

Editor's message: the sunk cost fallacy of deep drilling

At the beginning of 2008, I visited a watershed, located in Karkinatam village in the state of Karnataka, South India, where crops are intensively irrigated using groundwater. The water table had been depleted from a depth of 5 to 50 m in a large part of the area. Presently, 42% of a total of 158 water wells in the watershed are dry. Speaking with the farmers, I have been amazed to learn that they were drilling down to 500 m to tap water. This case is, of course, not isolated.

Deep Packet Inspection Using Message Passing Networks

We propose a solution based on message passing bipartite networks, for deep packet inspection, which addresses both speed and memory issues, which are limiting factors in current solutions. We report on a preliminary implementation and propose a parallel architecture.

Characterization of the interactions of a polycationic, amphiphilic, terminally branched oligopeptide with lipid A and lipopolysaccharide from the deep rough mutant of Salmonella minnesota

The lipid A and lipopolysaccharide (LPS) binding and neutralizing activities of a synthetic, polycationic, amphiphilic peptide were studied. The branched peptide, designed as a functional analog of polymyxin B, has a six residue hydrophobic sequence, bearing at its N-terminus a penultimate lysine residue whose alpha- and epsilon-amino groups are coupled to two terminal lysine residues. In fluorescence spectroscopic studies designed to examine relative affinities of binding to the toxin, neutralization of surface charge and fluidization of the acyl domains, the peptide was active, closely resembling the effects of polymyxin B and its nonapeptide derivative; however, the synthetic peptide does not induce phase transitions in LPS aggregates as do polymyxin B and polymyxin B nonapeptide. The peptide was also comparable with polymyxin B in its ability to inhibit LPS-mediated IL-l and IL-6 release from human peripheral blood mononuclear cells. The synthetic compound is devoid of antibacterial activities and did not induce conductance fluxes in LPS-containing asymmetric planar membranes. These results strengthen the premise that basicity and amphiphilicity are necessary and sufficient physical properties that ascribe endotoxin binding and neutralizing activities, and further suggest that antibacterial/membrane perturbant and LPS neutralizing activities are dissociable, which may be of value in designing LPS-sequestering agents of low toxicity.

Gravitational stresses in single span deep beams

A theoretical solution for the gravitational stresses in single span deep beams using Fourier series has been given. Numerical results for different span to depth ratios are given and these have been compared with the photoelastic results given by Saad and Hendry [1], and the finite difference results of Chow et al. [2,3].