Models, code, and papers for "Saurabh Singh":
Batch normalization (BN) has been very effective for deep learning and is widely used. However, when training with small minibatches, models using BN exhibit a significant degradation in performance. In this paper we study this peculiar behavior of BN to gain a better understanding of the problem, and identify a potential cause based on a statistical insight. We propose `EvalNorm' to address the issue by estimating corrected normalization statistics to use for BN during evaluation. EvalNorm supports online estimation of the corrected statistics while the model is being trained, and it does not affect the training scheme of the model. As a result, an added advantage of EvalNorm is that it can be used with existing pre-trained models allowing them to benefit from our method. EvalNorm yields large gains for models trained with smaller batches. Our experiments show that EvalNorm performs 6.18% (absolute) better than vanilla BN for a batchsize of 2 on ImageNet validation set and from 1.5 to 7.0 points (absolute) gain on the COCO object detection benchmark across a variety of setups.
We describe Swapout, a new stochastic training method, that outperforms ResNets of identical network structure yielding impressive results on CIFAR-10 and CIFAR-100. Swapout samples from a rich set of architectures including dropout, stochastic depth and residual architectures as special cases. When viewed as a regularization method swapout not only inhibits co-adaptation of units in a layer, similar to dropout, but also across network layers. We conjecture that swapout achieves strong regularization by implicitly tying the parameters across layers. When viewed as an ensemble training method, it samples a much richer set of architectures than existing methods such as dropout or stochastic depth. We propose a parameterization that reveals connections to exiting architectures and suggests a much richer set of architectures to be explored. We show that our formulation suggests an efficient training method and validate our conclusions on CIFAR-10 and CIFAR-100 matching state of the art accuracy. Remarkably, our 32 layer wider model performs similar to a 1001 layer ResNet model.
The goal of this paper is to discover a set of discriminative patches which can serve as a fully unsupervised mid-level visual representation. The desired patches need to satisfy two requirements: 1) to be representative, they need to occur frequently enough in the visual world; 2) to be discriminative, they need to be different enough from the rest of the visual world. The patches could correspond to parts, objects, "visual phrases", etc. but are not restricted to be any one of them. We pose this as an unsupervised discriminative clustering problem on a huge dataset of image patches. We use an iterative procedure which alternates between clustering and training discriminative classifiers, while applying careful cross-validation at each step to prevent overfitting. The paper experimentally demonstrates the effectiveness of discriminative patches as an unsupervised mid-level visual representation, suggesting that it could be used in place of visual words for many tasks. Furthermore, discriminative patches can also be used in a supervised regime, such as scene classification, where they demonstrate state-of-the-art performance on the MIT Indoor-67 dataset.
The success of deep learning in vision can be attributed to: (a) models with high capacity; (b) increased computational power; and (c) availability of large-scale labeled data. Since 2012, there have been significant advances in representation capabilities of the models and computational capabilities of GPUs. But the size of the biggest dataset has surprisingly remained constant. What will happen if we increase the dataset size by 10x or 100x? This paper takes a step towards clearing the clouds of mystery surrounding the relationship between `enormous data' and visual deep learning. By exploiting the JFT-300M dataset which has more than 375M noisy labels for 300M images, we investigate how the performance of current vision tasks would change if this data was used for representation learning. Our paper delivers some surprising (and some expected) findings. First, we find that the performance on vision tasks increases logarithmically based on volume of training data size. Second, we show that representation learning (or pre-training) still holds a lot of promise. One can improve performance on many vision tasks by just training a better base model. Finally, as expected, we present new state-of-the-art results for different vision tasks including image classification, object detection, semantic segmentation and human pose estimation. Our sincere hope is that this inspires vision community to not undervalue the data and develop collective efforts in building larger datasets.
We present a method that learns to answer visual questions by selecting image regions relevant to the text-based query. Our method exhibits significant improvements in answering questions such as "what color," where it is necessary to evaluate a specific location, and "what room," where it selectively identifies informative image regions. Our model is tested on the VQA dataset which is the largest human-annotated visual question answering dataset to our knowledge.
We describe an end-to-end neural network weight compression approach that draws inspiration from recent latent-variable data compression methods. The network parameters (weights and biases) are represented in a "latent" space, amounting to a reparameterization. This space is equipped with a learned probability model, which is used to impose an entropy penalty on the parameter representation during training, and to compress the representation using arithmetic coding after training. We are thus maximizing accuracy and model compressibility jointly, in an end-to-end fashion, with the rate--error trade-off specified by a hyperparameter. We evaluate our method by compressing six distinct model architectures on the MNIST, CIFAR-10 and ImageNet classification benchmarks. Our method achieves state-of-the-art compression on VGG-16, LeNet300-100 and several ResNet architectures, and is competitive on LeNet-5.
An important goal of computer vision is to build systems that learn visual representations over time that can be applied to many tasks. In this paper, we investigate a vision-language embedding as a core representation and show that it leads to better cross-task transfer than standard multi-task learning. In particular, the task of visual recognition is aligned to the task of visual question answering by forcing each to use the same word-region embeddings. We show this leads to greater inductive transfer from recognition to VQA than standard multitask learning. Visual recognition also improves, especially for categories that have relatively few recognition training labels but appear often in the VQA setting. Thus, our paper takes a small step towards creating more general vision systems by showing the benefit of interpretable, flexible, and trainable core representations.
User-given tags or labels are valuable resources for semantic understanding of visual media such as images and videos. Recently, a new type of labeling mechanism known as hash-tags have become increasingly popular on social media sites. In this paper, we study the problem of generating relevant and useful hash-tags for short video clips. Traditional data-driven approaches for tag enrichment and recommendation use direct visual similarity for label transfer and propagation. We attempt to learn a direct low-cost mapping from video to hash-tags using a two step training process. We first employ a natural language processing (NLP) technique, skip-gram models with neural network training to learn a low-dimensional vector representation of hash-tags (Tag2Vec) using a corpus of 10 million hash-tags. We then train an embedding function to map video features to the low-dimensional Tag2vec space. We learn this embedding for 29 categories of short video clips with hash-tags. A query video without any tag-information can then be directly mapped to the vector space of tags using the learned embedding and relevant tags can be found by performing a simple nearest-neighbor retrieval in the Tag2Vec space. We validate the relevance of the tags suggested by our system qualitatively and quantitatively with a user study.
Short internet video clips like vines present a significantly wild distribution compared to traditional video datasets. In this paper, we focus on the problem of unsupervised action classification in wild vines using traditional labeled datasets. To this end, we use a data augmentation based simple domain adaptation strategy. We utilise semantic word2vec space as a common subspace to embed video features from both, labeled source domain and unlablled target domain. Our method incrementally augments the labeled source with target samples and iteratively modifies the embedding function to bring the source and target distributions together. Additionally, we utilise a multi-modal representation that incorporates noisy semantic information available in form of hash-tags. We show the effectiveness of this simple adaptation technique on a test set of vines and achieve notable improvements in performance.
We present a simple deep learning framework to simultaneously predict keypoint locations and their respective visibilities and use those to achieve state-of-the-art performance for fine-grained classification. We show that by conditioning the predictions on object proposals with sufficient image support, our method can do well without complicated spatial reasoning. Instead, inference methods with robustness to outliers, yield state-of-the-art for keypoint localization. We demonstrate the effectiveness of our accurate keypoint localization and visibility prediction on the fine-grained bird recognition task with and without ground truth bird bounding boxes, and outperform existing state-of-the-art methods by over 2%.
Effective SLAM using a single monocular camera is highly preferred due to its simplicity. However, when compared to trajectory planning methods using depth-based SLAM, Monocular SLAM in loop does need additional considerations. One main reason being that for the optimization, in the form of Bundle Adjustment (BA), to be robust, the SLAM system needs to scan the area for a reasonable duration. Most monocular SLAM systems do not tolerate large camera rotations between successive views and tend to breakdown. Other reasons for Monocular SLAM failure include ambiguities in decomposition of the Essential Matrix, feature-sparse scenes and more layers of non linear optimization apart from BA. This paper presents a novel formulation based on Reinforcement Learning (RL) that generates fail safe trajectories wherein the SLAM generated outputs (scene structure and camera motion) do not deviate largely from their true values. Quintessentially, the RL framework successfully learns the otherwise complex relation between motor actions and perceptual inputs that result in trajectories that do not cause failure of SLAM, which are almost intractable to capture in an obvious mathematical formulation. We show systematically in simulations how the quality of the SLAM map and trajectory dramatically improves when trajectories are computed by using RL.
The leading approach for image compression with artificial neural networks (ANNs) is to learn a nonlinear transform and a fixed entropy model that are optimized for rate-distortion performance. We show that this approach can be significantly improved by incorporating spatially local, image-dependent entropy models. The key insight is that existing ANN-based methods learn an entropy model that is shared between the encoder and decoder, but they do not transmit any side information that would allow the model to adapt to the structure of a specific image. We present a method for augmenting ANN-based image coders with image-dependent side information that leads to a 17.8% rate reduction over a state-of-the-art ANN-based baseline model on a standard evaluation set, and 70-98% reductions on images with low visual complexity that are poorly captured by a fixed, global entropy model.
Code-mixing or code-switching are the effortless phenomena of natural switching between two or more languages in a single conversation. Use of a foreign word in a language; however, does not necessarily mean that the speaker is code-switching because often languages borrow lexical items from other languages. If a word is borrowed, it becomes a part of the lexicon of a language; whereas, during code-switching, the speaker is aware that the conversation involves foreign words or phrases. Identifying whether a foreign word used by a bilingual speaker is due to borrowing or code-switching is a fundamental importance to theories of multilingualism, and an essential prerequisite towards the development of language and speech technologies for multilingual communities. In this paper, we present a series of novel computational methods to identify the borrowed likeliness of a word, based on the social media signals. We first propose context based clustering method to sample a set of candidate words from the social media data.Next, we propose three novel and similar metrics based on the usage of these words by the users in different tweets; these metrics were used to score and rank the candidate words indicating their borrowed likeliness. We compare these rankings with a ground truth ranking constructed through a human judgment experiment. The Spearman's rank correlation between the two rankings (nearly 0.62 for all the three metric variants) is more than double the value (0.26) of the most competitive existing baseline reported in the literature. Some other striking observations are, (i) the correlation is higher for the ground truth data elicited from the younger participants (age less than 30) than that from the older participants, and (ii )those participants who use mixed-language for tweeting the least, provide the best signals of borrowing.
We address the problem of distance metric learning (DML), defined as learning a distance consistent with a notion of semantic similarity. Traditionally, for this problem supervision is expressed in the form of sets of points that follow an ordinal relationship -- an anchor point $x$ is similar to a set of positive points $Y$, and dissimilar to a set of negative points $Z$, and a loss defined over these distances is minimized. While the specifics of the optimization differ, in this work we collectively call this type of supervision Triplets and all methods that follow this pattern Triplet-Based methods. These methods are challenging to optimize. A main issue is the need for finding informative triplets, which is usually achieved by a variety of tricks such as increasing the batch size, hard or semi-hard triplet mining, etc. Even with these tricks, the convergence rate of such methods is slow. In this paper we propose to optimize the triplet loss on a different space of triplets, consisting of an anchor data point and similar and dissimilar proxy points which are learned as well. These proxies approximate the original data points, so that a triplet loss over the proxies is a tight upper bound of the original loss. This proxy-based loss is empirically better behaved. As a result, the proxy-loss improves on state-of-art results for three standard zero-shot learning datasets, by up to 15% points, while converging three times as fast as other triplet-based losses.
In this paper, we present a set of computational methods to identify the likeliness of a word being borrowed, based on the signals from social media. In terms of Spearman correlation coefficient values, our methods perform more than two times better (nearly 0.62) in predicting the borrowing likeliness compared to the best performing baseline (nearly 0.26) reported in literature. Based on this likeliness estimate we asked annotators to re-annotate the language tags of foreign words in predominantly native contexts. In 88 percent of cases the annotators felt that the foreign language tag should be replaced by native language tag, thus indicating a huge scope for improvement of automatic language identification systems.
Many real world systems need to operate on heterogeneous information networks that consist of numerous interacting components of different types. Examples include systems that perform data analysis on biological information networks; social networks; and information extraction systems processing unstructured data to convert raw text to knowledge graphs. Many previous works describe specialized approaches to perform specific types of analysis, mining and learning on such networks. In this work, we propose a unified framework consisting of a data model -a graph with a first order schema along with a declarative language for constructing, querying and manipulating such networks in ways that facilitate relational and structured machine learning. In particular, we provide an initial prototype for a relational and graph traversal query language where queries are directly used as relational features for structured machine learning models. Feature extraction is performed by making declarative graph traversal queries. Learning and inference models can directly operate on this relational representation and augment it with new data and knowledge that, in turn, is integrated seamlessly into the relational structure to support new predictions. We demonstrate this system's capabilities by showcasing tasks in natural language processing and computational biology domains.
We introduce a stop-code tolerant (SCT) approach to training recurrent convolutional neural networks for lossy image compression. Our methods introduce a multi-pass training method to combine the training goals of high-quality reconstructions in areas around stop-code masking as well as in highly-detailed areas. These methods lead to lower true bitrates for a given recursion count, both pre- and post-entropy coding, even using unstructured LZ77 code compression. The pre-LZ77 gains are achieved by trimming stop codes. The post-LZ77 gains are due to the highly unequal distributions of 0/1 codes from the SCT architectures. With these code compressions, the SCT architecture maintains or exceeds the image quality at all compression rates compared to JPEG and to RNN auto-encoders across the Kodak dataset. In addition, the SCT coding results in lower variance in image quality across the extent of the image, a characteristic that has been shown to be important in human ratings of image quality
Monocular SLAM refers to using a single camera to estimate robot ego motion while building a map of the environment. While Monocular SLAM is a well studied problem, automating Monocular SLAM by integrating it with trajectory planning frameworks is particularly challenging. This paper presents a novel formulation based on Reinforcement Learning (RL) that generates fail safe trajectories wherein the SLAM generated outputs do not deviate largely from their true values. Quintessentially, the RL framework successfully learns the otherwise complex relation between perceptual inputs and motor actions and uses this knowledge to generate trajectories that do not cause failure of SLAM. We show systematically in simulations how the quality of the SLAM dramatically improves when trajectories are computed using RL. Our method scales effectively across Monocular SLAM frameworks in both simulation and in real world experiments with a mobile robot.
Deep neural networks represent a powerful class of function approximators that can learn to compress and reconstruct images. Existing image compression algorithms based on neural networks learn quantized representations with a constant spatial bit rate across each image. While entropy coding introduces some spatial variation, traditional codecs have benefited significantly by explicitly adapting the bit rate based on local image complexity and visual saliency. This paper introduces an algorithm that combines deep neural networks with quality-sensitive bit rate adaptation using a tiled network. We demonstrate the importance of spatial context prediction and show improved quantitative (PSNR) and qualitative (subjective rater assessment) results compared to a non-adaptive baseline and a recently published image compression model based on fully-convolutional neural networks.
We propose a method for lossy image compression based on recurrent, convolutional neural networks that outperforms BPG (4:2:0 ), WebP, JPEG2000, and JPEG as measured by MS-SSIM. We introduce three improvements over previous research that lead to this state-of-the-art result. First, we show that training with a pixel-wise loss weighted by SSIM increases reconstruction quality according to several metrics. Second, we modify the recurrent architecture to improve spatial diffusion, which allows the network to more effectively capture and propagate image information through the network's hidden state. Finally, in addition to lossless entropy coding, we use a spatially adaptive bit allocation algorithm to more efficiently use the limited number of bits to encode visually complex image regions. We evaluate our method on the Kodak and Tecnick image sets and compare against standard codecs as well recently published methods based on deep neural networks.