arXiv subjects

Ramesh Raskar

Publications and source records attributed to Ramesh Raskar.

At least 127 records · Page 7Linked to original sources

arXiv2018

Flash Photography for Data-Driven Hidden Scene Recovery

Vehicles, search and rescue personnel, and endoscopes use flash lights to locate, identify, and view objects in their surroundings. Here we show the first steps of how all these tasks can be done around corners with consumer cameras. Recent techniques for NLOS imaging using consumer cameras have not been able to both localize and identify the hidden object. We introduce a method that couples traditional geometric understanding and data-driven techniques. To avoid the limitation of large dataset gathering, we train the data-driven models on rendered samples to computationally recover the hidden scene on real data. The method has three independent operating modes: 1) a regression output to localize a hidden object in 2D, 2) an identification output to identify the object type or pose, and 3) a generative network to reconstruct the hidden scene from a new viewpoint. The method is able to localize 12cm wide hidden objects in 2D with 1.7cm accuracy. The method also identifies the hidden object class with 87.7% accuracy (compared to 33.3% random accuracy). This paper also provides an analysis on the distribution of information that encodes the occluded object in the accessible scene. We show that, unlike previously thought, the area that extends beyond the corner is essential for accurate object localization and identification.

cs.CV↗

arXiv2018

Distributed learning of deep neural network over multiple agents

In domains such as health care and finance, shortage of labeled data and computational resources is a critical issue while developing machine learning algorithms. To address the issue of labeled data scarcity in training and deployment of neural network-based systems, we propose a new technique to train deep neural networks over several data sources. Our method allows for deep neural networks to be trained using data from multiple entities in a distributed fashion. We evaluate our algorithm on existing datasets and show that it obtains performance which is similar to a regular neural network trained on a single machine. We further extend it to incorporate semi-supervised learning when training with few labeled samples, and analyze any security concerns that may arise. Our algorithm paves the way for distributed training of deep neural networks in data sensitive applications when raw data may not be shared directly.

cs.LG↗

arXiv2018

Maximum-Entropy Fine-Grained Classification

Fine-Grained Visual Classification (FGVC) is an important computer vision problem that involves small diversity within the different classes, and often requires expert annotators to collect data. Utilizing this notion of small visual diversity, we revisit Maximum-Entropy learning in the context of fine-grained classification, and provide a training routine that maximizes the entropy of the output probability distribution for training convolutional neural networks on FGVC tasks. We provide a theoretical as well as empirical justification of our approach, and achieve state-of-the-art performance across a variety of classification tasks in FGVC, that can potentially be extended to any fine-tuning task. Our method is robust to different hyperparameter values, amount of training data and amount of training label noise and can hence be a valuable tool in many similar problems.

cs.CV↗

arXiv2018

Pairwise Confusion for Fine-Grained Visual Classification

Fine-Grained Visual Classification (FGVC) datasets contain small sample sizes, along with significant intra-class variation and inter-class similarity. While prior work has addressed intra-class variation using localization and segmentation techniques, inter-class similarity may also affect feature learning and reduce classification performance. In this work, we address this problem using a novel optimization procedure for the end-to-end neural network training on FGVC tasks. Our procedure, called Pairwise Confusion (PC) reduces overfitting by intentionally {introducing confusion} in the activations. With PC regularization, we obtain state-of-the-art performance on six of the most widely-used FGVC datasets and demonstrate improved localization ability. {PC} is easy to implement, does not need excessive hyperparameter tuning during training, and does not add significant overhead during test time.

cs.CV↗

arXiv2018

DeepGlobe 2018: A Challenge to Parse the Earth through Satellite Images

We present the DeepGlobe 2018 Satellite Image Understanding Challenge, which includes three public competitions for segmentation, detection, and classification tasks on satellite images. Similar to other challenges in computer vision domain such as DAVIS and COCO, DeepGlobe proposes three datasets and corresponding evaluation methodologies, coherently bundled in three competitions with a dedicated workshop co-located with CVPR 2018. We observed that satellite imagery is a rich and structured source of information, yet it is less investigated than everyday images by computer vision researchers. However, bridging modern computer vision with remote sensing data analysis could have critical impact to the way we understand our environment and lead to major breakthroughs in global urban planning or climate change research. Keeping such bridging objective in mind, DeepGlobe aims to bring together researchers from different domains to raise awareness of remote sensing in the computer vision community and vice-versa. We aim to improve and evaluate state-of-the-art satellite image understanding approaches, which can hopefully serve as reference benchmarks for future research in the same topic. In this paper, we analyze characteristics of each dataset, define the evaluation criteria of the competitions, and provide baselines for each task.

cs.CV↗

arXiv2018

Economic Impact of Discoverability of Localities and Addresses in India

Most of the earth's population has a poorly defined addressing system, thus having a poorly discoverable residence, property or business locations on a map. Easily discoverable addresses are important for improving their livelihood, business-incomes, and even service delivery. The economic cycle based on discoverable addresses is self-reinforcing: consumers independently identify and adopt such addresses according to their own convenience and businesses use algorithms or third-party services to resolve these addresses into geocodes to help better identify their customers' locations. Our paper analyses from the top two industries in India: Logistics and Financial Services, indicate that the lack of a good addressing system costs India $10-14B annually. As the Indian economy is expected to grow rapidly, the businesses would proportionately grow, causing the total costs to grow further. We, therefore, need to consider a dramatically new approach to modernize the addressing systems to bring in efficiency.

cs.CY↗

arXiv2018

What is the right addressing scheme for India?

Computer generated addresses are coming to your neighborhood because most places in the world do not have an assigned meaningful street address. In India, 80% of the addresses are written with respect to a landmark which typically lies between 50-1500 meters of the actual address; such addresses make geolocating very challenging. Accuracy in geolocation is critical for emergency services to navigate quickly to reach you and for logistics industries to improve on-time performance and efficient routing of the package coming to your house. In this paper, we explore suggested addressing schemes for India, to determine what use cases and potential technologies will have the best adoption and therefore, greatest impact.

cs.CY↗

arXiv2017

Accelerating Neural Architecture Search using Performance Prediction

Methods for neural network hyperparameter optimization and meta-modeling are computationally expensive due to the need to train a large number of model configurations. In this paper, we show that standard frequentist regression models can predict the final performance of partially trained model configurations using features based on network architectures, hyperparameters, and time-series validation performance data. We empirically show that our performance prediction models are much more effective than prominent Bayesian counterparts, are simpler to implement, and are faster to train. Our models can predict final performance in both visual classification and language modeling domains, are effective for predicting performance of drastically varying model architectures, and can even generalize between model classes. Using these prediction models, we also propose an early stopping method for hyperparameter optimization and meta-modeling, which obtains a speedup of a factor up to 6x in both hyperparameter optimization and meta-modeling. Finally, we empirically show that our early stopping method can be seamlessly incorporated into both reinforcement learning-based architecture selection algorithms and bandit based search methods. Through extensive experimentation, we empirically show our performance prediction models and early stopping algorithm are state-of-the-art in terms of prediction accuracy and speedup achieved while still identifying the optimal model configurations.

cs.LG↗

arXiv2017

On Unlimited Sampling

Shannon's sampling theorem provides a link between the continuous and the discrete realms stating that bandlimited signals are uniquely determined by its values on a discrete set. This theorem is realized in practice using so called analog--to--digital converters (ADCs). Unlike Shannon's sampling theorem, the ADCs are limited in dynamic range. Whenever a signal exceeds some preset threshold, the ADC saturates, resulting in aliasing due to clipping. The goal of this paper is to analyze an alternative approach that does not suffer from these problems. Our work is based on recent developments in ADC design, which allow for ADCs that reset rather than to saturate, thus producing modulo samples. An open problem that remains is: Given such modulo samples of a bandlimited function as well as the dynamic range of the ADC, how can the original signal be recovered and what are the sufficient conditions that guarantee perfect recovery? In this paper, we prove such sufficiency conditions and complement them with a stable recovery algorithm. Our results are not limited to certain amplitude ranges, in fact even the same circuit architecture allows for the recovery of arbitrary large amplitudes as long as some estimate of the signal norm is available when recovering. Numerical experiments that corroborate our theory indeed show that it is possible to perfectly recover function that takes values that are orders of magnitude higher than the ADC's threshold.

cs.IT↗

arXiv2017

Lensless Imaging with Compressive Ultrafast Sensing

Lensless imaging is an important and challenging problem. One notable solution to lensless imaging is a single pixel camera which benefits from ideas central to compressive sampling. However, traditional single pixel cameras require many illumination patterns which result in a long acquisition process. Here we present a method for lensless imaging based on compressive ultrafast sensing. Each sensor acquisition is encoded with a different illumination pattern and produces a time series where time is a function of the photon's origin in the scene. Currently available hardware with picosecond time resolution enables time tagging photons as they arrive to an omnidirectional sensor. This allows lensless imaging with significantly fewer patterns compared to regular single pixel imaging. To that end, we develop a framework for designing lensless imaging systems that use ultrafast detectors. We provide an algorithm for ideal sensor placement and an algorithm for optimized active illumination patterns. We show that efficient lensless imaging is possible with ultrafast measurement and compressive sensing. This paves the way for novel imaging architectures and remote sensing in extreme situations where imaging with a lens is not possible.

cs.CV↗

arXiv2017

Designing Neural Network Architectures using Reinforcement Learning

At present, designing convolutional neural network (CNN) architectures requires both human expertise and labor. New architectures are handcrafted by careful experimentation or modified from a handful of existing networks. We introduce MetaQNN, a meta-modeling algorithm based on reinforcement learning to automatically generate high-performing CNN architectures for a given learning task. The learning agent is trained to sequentially choose CNN layers using $Q$-learning with an $ε$-greedy exploration strategy and experience replay. The agent explores a large but finite space of possible architectures and iteratively discovers designs with improved performance on the learning task. On image classification benchmarks, the agent-designed networks (consisting of only standard convolution, pooling, and fully-connected layers) beat existing networks designed with the same layer types and are competitive against the state-of-the-art methods that use more complex layer types. We also outperform existing meta-modeling approaches for network design on image classification tasks.

cs.LG↗

arXiv2017

Sampling Without Time: Recovering Echoes of Light via Temporal Phase Retrieval

This paper considers the problem of sampling and reconstruction of a continuous-time sparse signal without assuming the knowledge of the sampling instants or the sampling rate. This topic has its roots in the problem of recovering multiple echoes of light from its low-pass filtered and auto-correlated, time-domain measurements. Our work is closely related to the topic of sparse phase retrieval and in this context, we discuss the advantage of phase-free measurements. While this problem is ill-posed, cues based on physical constraints allow for its appropriate regularization. We validate our theory with experiments based on customized, optical time-of-flight imaging sensors. What singles out our approach is that our sensing method allows for temporal phase retrieval as opposed to the usual case of spatial phase retrieval. Preliminary experiments and results demonstrate a compelling capability of our phase-retrieval based imaging device.

cs.IT↗

arXiv2016

Deep Learning the City : Quantifying Urban Perception At A Global Scale

Computer vision methods that quantify the perception of urban environment are increasingly being used to study the relationship between a city's physical appearance and the behavior and health of its residents. Yet, the throughput of current methods is too limited to quantify the perception of cities across the world. To tackle this challenge, we introduce a new crowdsourced dataset containing 110,988 images from 56 cities, and 1,170,000 pairwise comparisons provided by 81,630 online volunteers along six perceptual attributes: safe, lively, boring, wealthy, depressing, and beautiful. Using this data, we train a Siamese-like convolutional neural architecture, which learns from a joint classification and ranking loss, to predict human judgments of pairwise image comparisons. Our results show that crowdsourcing combined with neural networks can produce urban perception data at the global scale.

cs.CV↗

arXiv2016

Shape from Mixed Polarization

Shape from Polarization (SfP) estimates surface normals using photos captured at different polarizer rotations. Fundamentally, the SfP model assumes that light is reflected either diffusely or specularly. However, this model is not valid for many real-world surfaces exhibiting a mixture of diffuse and specular properties. To address this challenge, previous methods have used a sequential solution: first, use an existing algorithm to separate the scene into diffuse and specular components, then apply the appropriate SfP model. In this paper, we propose a new method that jointly uses viewpoint and polarization data to holistically separate diffuse and specular components, recover refractive index, and ultimately recover 3D shape. By involving the physics of polarization in the separation process, we demonstrate competitive results with a benchmark method, while recovering additional information (e.g. refractive index).

cs.CV↗

arXiv2016

Sweep Distortion Removal from THz Images via Blind Demodulation

Heavy sweep distortion induced by alignments and inter-reflections of layers of a sample is a major burden in recovering 2D and 3D information in time resolved spectral imaging. This problem cannot be addressed by conventional denoising and signal processing techniques as it heavily depends on the physics of the acquisition. Here we propose and implement an algorithmic framework based on low-rank matrix recovery and alternating minimization that exploits the forward model for THz acquisition. The method allows recovering the original signal in spite of the presence of temporal-spatial distortions. We address a blind-demodulation problem, where based on several observations of the sample texture modulated by an undesired sweep pattern, the two classes of signals are separated. The performance of the method is examined in both synthetic and experimental data, and the successful reconstructions are demonstrated. The proposed general scheme can be implemented to advance inspection and imaging applications in THz and other time-resolved sensing modalities.

cs.CV↗

arXiv2016

Multi-velocity neural networks for gesture recognition in videos

We present a new action recognition deep neural network which adaptively learns the best action velocities in addition to the classification. While deep neural networks have reached maturity for image understanding tasks, we are still exploring network topologies and features to handle the richer environment of video clips. Here, we tackle the problem of multiple velocities in action recognition, and provide state-of-the-art results for gesture recognition, on known and new collected datasets. We further provide the training steps for our semi-supervised network, suited to learn from huge unlabeled datasets with only a fraction of labeled examples.

cs.CV↗

arXiv2016

Deep video gesture recognition using illumination invariants

In this paper we present architectures based on deep neural nets for gesture recognition in videos, which are invariant to local scaling. We amalgamate autoencoder and predictor architectures using an adaptive weighting scheme coping with a reduced size labeled dataset, while enriching our models from enormous unlabeled sets. We further improve robustness to lighting conditions by introducing a new adaptive filer based on temporal local scale normalization. We provide superior results over known methods, including recent reported approaches based on neural nets.

cs.CV↗

arXiv2015

Coreset-Based Adaptive Tracking

We propose a method for learning from streaming visual data using a compact, constant size representation of all the data that was seen until a given moment. Specifically, we construct a 'coreset' representation of streaming data using a parallelized algorithm, which is an approximation of a set with relation to the squared distances between this set and all other points in its ambient space. We learn an adaptive object appearance model from the coreset tree in constant time and logarithmic space and use it for object tracking by detection. Our method obtains excellent results for object tracking on three standard datasets over more than 100 videos. The ability to summarize data efficiently makes our method ideally suited for tracking in long videos in presence of space and time constraints. We demonstrate this ability by outperforming a variety of algorithms on the TLD dataset with 2685 frames on average. This coreset based learning approach can be applied for both real-time learning of small, varied data and fast learning of big data.

cs.CV↗