Deep neural networks (DNNs) demand a very large amount of computation and weight storage, and thus efficient implementation using special purpose hardware is highly desired. In this work, we have developed an FPGA based fixed-point DNN system using only on-chip memory not to access external DRAM. The execution time and energy consumption of the developed system is compared with a GPU based implementation. Since the capacity of memory in FPGA is limited, only 3-bit weights are used for this implementation, and training based fixed-point weight optimization is employed. The implementation using Xilinx XC7Z045 is tested for the MNIST handwritten digit recognition benchmark and a phoneme recognition task on TIMIT corpus. The obtained speed is about one quarter of a GPU based implementation and much better than that of a PC based one. The power consumption is less than 5 Watt at the full speed operation resulting in much higher efficiency compared to GPU based systems.
We present an information-theoretic framework for solving global black-box optimization problems that also have black-box constraints. Of particular interest to us is to efficiently solve problems with decoupled constraints, in which subsets of the objective and constraint functions may be evaluated independently. For example, when the objective is evaluated on a CPU and the constraints are evaluated independently on a GPU. These problems require an acquisition function that can be separated into the contributions of the individual function evaluations. We develop one such acquisition function and call it Predictive Entropy Search with Constraints (PESC). PESC is an approximation to the expected information gain criterion and it compares favorably to alternative approaches based on improvement in several synthetic and real-world problems. In addition to this, we consider problems with a mix of functions that are fast and slow to evaluate. These problems require balancing the amount of time spent in the meta-computation of PESC and in the actual evaluation of the target objective. We take a bounded rationality approach and develop partial update for PESC which trades off accuracy against speed. We then propose a method for adaptively switching between the partial and full updates for PESC. This allows us to interpolate between versions of PESC that are efficient in terms of function evaluations and those that are efficient in terms of wall-clock time. Overall, we demonstrate that PESC is an effective algorithm that provides a promising direction towards a unified solution for constrained Bayesian optimization.
In this work we introduce a convolutional neural network (CNN) that jointly handles low-, mid-, and high-level vision tasks in a unified architecture that is trained end-to-end. Such a universal network can act like a `swiss knife' for vision tasks; we call this architecture an UberNet to indicate its overarching nature. We address two main technical challenges that emerge when broadening up the range of tasks handled by a single CNN: (i) training a deep architecture while relying on diverse training sets and (ii) training many (potentially unlimited) tasks with a limited memory budget. Properly addressing these two problems allows us to train accurate predictors for a host of tasks, without compromising accuracy. Through these advances we train in an end-to-end manner a CNN that simultaneously addresses (a) boundary detection (b) normal estimation (c) saliency estimation (d) semantic segmentation (e) human part segmentation (f) semantic boundary detection, (g) region proposal generation and object detection. We obtain competitive performance while jointly addressing all of these tasks in 0.7 seconds per frame on a single GPU. A demonstration of this system can be found at http://cvn.ecp.fr/ubernet/.
Magnetic resonance imaging (MRI) nowadays serves as an important modality for diagnostic and therapeutic guidance in clinics. However, the {\it slow acquisition} process, the dynamic deformation of organs, as well as the need for {\it real-time} reconstruction, pose major challenges toward obtaining artifact-free images. To cope with these challenges, the present paper advocates a novel subspace learning framework that permeates benefits from parallel factor (PARAFAC) decomposition of tensors (multiway data) to low-rank modeling of temporal sequence of images. Treating images as multiway data arrays, the novel method preserves spatial structures and unravels the latent correlations across various dimensions by means of the tensor subspace. Leveraging the spatio-temporal correlation of images, Tykhonov regularization is adopted as a rank surrogate for a least-squares optimization program. Alteranating majorization minimization is adopted to develop online algorithms that recursively procure the reconstruction upon arrival of a new undersampled $k$-space frame. The developed algorithms are {\it provably convergent} and highly {\it parallelizable} with lightweight FFT tasks per iteration. To further accelerate the acquisition process, randomized subsampling policies are devised that leverage intermediate estimates of the tensor subspace, offered by the online scheme, to {\it randomly} acquire {\it informative} $k$-space samples. In a nutshell, the novel approach enables tracking motion dynamics under low acquisition rates `on the fly.' GPU-based tests with real {\it in vivo} MRI datasets of cardiac cine images corroborate the merits of the novel approach relative to state-of-the-art alternatives.
Accuracy, descriptor size, and the time required for extraction and matching are all important factors when selecting local image descriptors. To optimize over all these requirements, this paper presents a CUDA port for the recent Learned Arrangement of Three Patches (LATCH) binary descriptors to the GPU platform. The design of LATCH makes it well suited for GPU processing. Owing to its small size and binary nature, the GPU can further be used to efficiently match LATCH features. Taken together, this leads to breakneck descriptor extraction and matching speeds. We evaluate the trade off between these speeds and the quality of results in a feature matching intensive application. To this end, we use our proposed CUDA LATCH (CLATCH) to recover structure from motion (SfM), comparing 3D reconstructions and speed using different representations. Our results show that CLATCH provides high quality 3D reconstructions at fractions of the time required by other representations, with little, if any, loss of reconstruction quality.
Poisson denoising is an essential issue for various imaging applications, such as night vision, medical imaging and microscopy. State-of-the-art approaches are clearly dominated by patch-based non-local methods in recent years. In this paper, we aim to propose a local Poisson denoising model with both structure simplicity and good performance. To this end, we consider a variational modeling to integrate the so-called Fields of Experts (FoE) image prior, that has proven an effective higher-order Markov Random Fields (MRF) model for many classic image restoration problems. We exploit several feasible variational variants for this task. We start with a direct modeling in the original image domain by taking into account the Poisson noise statistics, which performs generally well for the cases of high SNR. However, this strategy encounters problem in cases of low SNR. Then we turn to an alternative modeling strategy by using the Anscombe transform and Gaussian statistics derived data term. We retrain the FoE prior model directly in the transform domain. With the newly trained FoE model, we end up with a local variational model providing strongly competitive results against state-of-the-art non-local approaches, meanwhile bearing the property of simple structure. Furthermore, our proposed model comes along with an additional advantage, that the inference is very efficient as it is well-suited for parallel computation on GPUs. For images of size $512 \times 512$, our GPU implementation takes less than 1 second to produce state-of-the-art Poisson denoising performance.
For applications in navigation and robotics, estimating the 3D pose of objects is as important as detection. Many approaches to pose estimation rely on detecting or tracking parts or keypoints [11, 21]. In this paper we build on a recent state-of-the-art convolutional network for slidingwindow detection [10] to provide detection and rough pose estimation in a single shot, without intermediate stages of detecting parts or initial bounding boxes. While not the first system to treat pose estimation as a categorization problem, this is the first attempt to combine detection and pose estimation at the same level using a deep learning approach. The key to the architecture is a deep convolutional network where scores for the presence of an object category, the offset for its location, and the approximate pose are all estimated on a regular grid of locations in the image. The resulting system is as accurate as recent work on pose estimation (42.4% 8 View mAVP on Pascal 3D+ [21] ) and significantly faster (46 frames per second (FPS) on a TITAN X GPU). This approach to detection and rough pose estimation is fast and accurate enough to be widely applied as a pre-processing step for tasks including high-accuracy pose estimation, object tracking and localization, and vSLAM.
We introduce a method to train Quantized Neural Networks (QNNs) --- neural networks with extremely low precision (e.g., 1-bit) weights and activations, at run-time. At train-time the quantized weights and activations are used for computing the parameter gradients. During the forward pass, QNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations. As a result, power consumption is expected to be drastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and ImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to their 32-bit counterparts. For example, our quantized version of AlexNet with 1-bit weights and 2-bit activations achieves $51\%$ top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients computation using only bit-wise operation. Quantized recurrent neural networks were tested over the Penn Treebank dataset, and achieved comparable accuracy as their 32-bit counterparts using only 4-bits. Last but not least, we programmed a binary matrix multiplication GPU kernel with which it is possible to run our MNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The QNN code is available online.
Recently, several models based on deep neural networks have achieved great success in terms of both reconstruction accuracy and computational performance for single image super-resolution. In these methods, the low resolution (LR) input image is upscaled to the high resolution (HR) space using a single filter, commonly bicubic interpolation, before reconstruction. This means that the super-resolution (SR) operation is performed in HR space. We demonstrate that this is sub-optimal and adds computational complexity. In this paper, we present the first convolutional neural network (CNN) capable of real-time SR of 1080p videos on a single K2 GPU. To achieve this, we propose a novel CNN architecture where the feature maps are extracted in the LR space. In addition, we introduce an efficient sub-pixel convolution layer which learns an array of upscaling filters to upscale the final LR feature maps into the HR output. By doing so, we effectively replace the handcrafted bicubic filter in the SR pipeline with more complex upscaling filters specifically trained for each feature map, whilst also reducing the computational complexity of the overall SR operation. We evaluate the proposed approach using images and videos from publicly available datasets and show that it performs significantly better (+0.15dB on Images and +0.39dB on Videos) and is an order of magnitude faster than previous CNN-based methods.
Deep learning has significantly advanced the state of the art in artificial intelligence, gaining wide popularity from both industry and academia. Special interest is around Convolutional Neural Networks (CNN), which take inspiration from the hierarchical structure of the visual cortex, to form deep layers of convolutional operations, along with fully connected classifiers. Hardware implementations of these deep CNN architectures are challenged with memory bottlenecks that require many convolution and fully-connected layers demanding large amount of communication for parallel computation. Multi-core CPU based solutions have demonstrated their inadequacy for this problem due to the memory wall and low parallelism. Many-core GPU architectures show superior performance but they consume high power and also have memory constraints due to inconsistencies between cache and main memory. FPGA design solutions are also actively being explored, which allow implementing the memory hierarchy using embedded BlockRAM. This boosts the parallel use of shared memory elements between multiple processing units, avoiding data replicability and inconsistencies. This makes FPGAs potentially powerful solutions for real-time classification of CNNs. Both Altera and Xilinx have adopted OpenCL co-design framework from GPU for FPGA designs as a pseudo-automatic development solution. In this paper, a comprehensive evaluation and comparison of Altera and Xilinx OpenCL frameworks for a 5-layer deep CNN is presented. Hardware resources, temporal performance and the OpenCL architecture for CNNs are discussed. Xilinx demonstrates faster synthesis, better FPGA resource utilization and more compact boards. Altera provides multi-platforms tools, mature design community and better execution times.
We propose an entirely data-driven approach to estimating the 3D pose of a hand given a depth image. We show that we can correct the mistakes made by a Convolutional Neural Network trained to predict an estimate of the 3D pose by using a feedback loop. The components of this feedback loop are also Deep Networks, optimized using training data. They remove the need for fitting a 3D model to the input data, which requires both a carefully designed fitting function and algorithm. We show that our approach outperforms state-of-the-art methods, and is efficient as our implementation runs at over 400 fps on a single GPU.
Convolutional Neural Networks (CNNs) have gained significant traction in the field of machine learning, particularly due to their high accuracy in visual recognition. Recent works have pushed the performance of GPU implementations of CNNs to significantly improve their classification and training times. With these improvements, many frameworks have become available for implementing CNNs on both CPUs and GPUs, with no support for FPGA implementations. In this work we present a modified version of the popular CNN framework Caffe, with FPGA support. This allows for classification using CNN models and specialized FPGA implementations with the flexibility of reprogramming the device when necessary, seamless memory transactions between host and device, simple-to-use test benches, and the ability to create pipelined layer implementations. To validate the framework, we use the Xilinx SDAccel environment to implement an FPGA-based Winograd convolution engine and show that the FPGA layer can be used alongside other layers running on a host processor to run several popular CNNs (AlexNet, GoogleNet, VGG A, Overfeat). The results show that our framework achieves 50 GFLOPS across 3x3 convolutions in the benchmarks. This is achieved within a practical framework, which will aid in future development of FPGA-based CNNs.
This paper presents how we can achieve the state-of-the-art accuracy in multi-category object detection task while minimizing the computational cost by adapting and combining recent technical innovations. Following the common pipeline of "CNN feature extraction + region proposal + RoI classification", we mainly redesign the feature extraction part, since region proposal part is not computationally expensive and classification part can be efficiently compressed with common techniques like truncated SVD. Our design principle is "less channels with more layers" and adoption of some building blocks including concatenated ReLU, Inception, and HyperNet. The designed network is deep and thin and trained with the help of batch normalization, residual connections, and learning rate scheduling based on plateau detection. We obtained solid results on well-known object detection benchmarks: 83.8% mAP (mean average precision) on VOC2007 and 82.5% mAP on VOC2012 (2nd place), while taking only 750ms/image on Intel i7-6700K CPU with a single core and 46ms/image on NVIDIA Titan X GPU. Theoretically, our network requires only 12.3% of the computational cost compared to ResNet-101, the winner on VOC2012.
Leveraging large data sets, deep Convolutional Neural Networks (CNNs) achieve state-of-the-art recognition accuracy. Due to the substantial compute and memory operations, however, they require significant execution time. The massive parallel computing capability of GPUs make them as one of the ideal platforms to accelerate CNNs and a number of GPU-based CNN libraries have been developed. While existing works mainly focus on the computational efficiency of CNNs, the memory efficiency of CNNs have been largely overlooked. Yet CNNs have intricate data structures and their memory behavior can have significant impact on the performance. In this work, we study the memory efficiency of various CNN layers and reveal the performance implication from both data layouts and memory access patterns. Experiments show the universal effect of our proposed optimizations on both single layers and various networks, with up to 27.9x for a single layer and up to 5.6x on the whole networks.
Latent Dirichlet Allocation (LDA) is a popular tool for analyzing discrete count data such as text and images. Applications require LDA to handle both large datasets and a large number of topics. Though distributed CPU systems have been used, GPU-based systems have emerged as a promising alternative because of the high computational power and memory bandwidth of GPUs. However, existing GPU-based LDA systems cannot support a large number of topics because they use algorithms on dense data structures whose time and space complexity is linear to the number of topics. In this paper, we propose SaberLDA, a GPU-based LDA system that implements a sparsity-aware algorithm to achieve sublinear time complexity and scales well to learn a large number of topics. To address the challenges introduced by sparsity, we propose a novel data layout, a new warp-based sampling kernel, and an efficient sparse count matrix updating algorithm that improves locality, makes efficient utilization of GPU warps, and reduces memory consumption. Experiments show that SaberLDA can learn from billions-token-scale data with up to 10,000 topics, which is almost two orders of magnitude larger than that of the previous GPU-based systems. With a single GPU card, SaberLDA is able to learn 10,000 topics from a dataset of billions of tokens in a few hours, which is only achievable with clusters with tens of machines before.
The Stixel World is a medium-level, compact representation of road scenes that abstracts millions of disparity pixels into hundreds or thousands of stixels. The goal of this work is to implement and evaluate a complete multi-stixel estimation pipeline on an embedded, energy-efficient, GPU-accelerated device. This work presents a full GPU-accelerated implementation of stixel estimation that produces reliable results at 26 frames per second (real-time) on the Tegra X1 for disparity images of 1024x440 pixels and stixel widths of 5 pixels, and achieves more than 400 frames per second on a high-end Titan X GPU card.
Dense, robust and real-time computation of depth information from stereo-camera systems is a computationally demanding requirement for robotics, advanced driver assistance systems (ADAS) and autonomous vehicles. Semi-Global Matching (SGM) is a widely used algorithm that propagates consistency constraints along several paths across the image. This work presents a real-time system producing reliable disparity estimation results on the new embedded energy-efficient GPU devices. Our design runs on a Tegra X1 at 42 frames per second (fps) for an image size of 640x480, 128 disparity levels, and using 4 path directions for the SGM method.
Many mobile applications running on smartphones and wearable devices would potentially benefit from the accuracy and scalability of deep CNN-based machine learning algorithms. However, performance and energy consumption limitations make the execution of such computationally intensive algorithms on mobile devices prohibitive. We present a GPU-accelerated library, dubbed CNNdroid, for execution of trained deep CNNs on Android-based mobile devices. Empirical evaluations show that CNNdroid achieves up to 60X speedup and 130X energy saving on current mobile devices. The CNNdroid open source library is available for download at https://github.com/ENCP/CNNdroid
High demand for computation resources severely hinders deployment of large-scale Deep Neural Networks (DNN) in resource constrained devices. In this work, we propose a Structured Sparsity Learning (SSL) method to regularize the structures (i.e., filters, channels, filter shapes, and layer depth) of DNNs. SSL can: (1) learn a compact structure from a bigger DNN to reduce computation cost; (2) obtain a hardware-friendly structured sparsity of DNN to efficiently accelerate the DNNs evaluation. Experimental results show that SSL achieves on average 5.1x and 3.1x speedups of convolutional layer computation of AlexNet against CPU and GPU, respectively, with off-the-shelf libraries. These speedups are about twice speedups of non-structured sparsity; (3) regularize the DNN structure to improve classification accuracy. The results show that for CIFAR-10, regularization on layer depth can reduce 20 layers of a Deep Residual Network (ResNet) to 18 layers while improve the accuracy from 91.25% to 92.60%, which is still slightly higher than that of original ResNet with 32 layers. For AlexNet, structure regularization by SSL also reduces the error by around ~1%. Open source code is in https://github.com/wenwei202/caffe/tree/scnn
We systematically explored a spectrum of normalization algorithms related to Batch Normalization (BN) and propose a generalized formulation that simultaneously solves two major limitations of BN: (1) online learning and (2) recurrent learning. Our proposal is simpler and more biologically-plausible. Unlike previous approaches, our technique can be applied out of the box to all learning scenarios (e.g., online learning, batch learning, fully-connected, convolutional, feedforward, recurrent and mixed --- recurrent and convolutional) and compare favorably with existing approaches. We also propose Lp Normalization for normalizing by different orders of statistical moments. In particular, L1 normalization is well-performing, simple to implement, fast to compute, more biologically-plausible and thus ideal for GPU or hardware implementations.