We present an overview of techniques for quantizing convolutional neural networks for inference with integer weights and activations. Per-channel quantization of weights and per-layer quantization of activations to 8-bits of precision post-training produces classification accuracies within 2% of floating point networks for a wide variety of CNN architectures. Model sizes can be reduced by a factor of 4 by quantizing weights to 8-bits, even when 8-bit arithmetic is not supported. This can be achieved with simple, post training quantization of weights.We benchmark latencies of quantized networks on CPUs and DSPs and observe a speedup of 2x-3x for quantized implementations compared to floating point on CPUs. Speedups of up to 10x are observed on specialized processors with fixed point SIMD capabilities, like the Qualcomm QDSPs with HVX. 
Quantization-aware training can provide further improvements, reducing the gap to floating point to 1% at 8-bit precision. Quantization-aware training also allows for reducing the precision of weights to four bits with accuracy losses ranging from 2% to 10%, with higher accuracy drop for smaller networks.We introduce tools in TensorFlow and TensorFlowLite for quantizing convolutional networks and review best practices for quantization-aware training to obtain high accuracy with quantized weights and activations. We recommend that per-channel quantization of weights and per-layer quantization of activations be the preferred quantization scheme for hardware acceleration and kernel optimization. We also propose that future processors and hardware accelerators for optimized inference support precisions of 4, 8 and 16 bits.

Quantizing deep convolutional networks for efficient inference: A whitepaper

Over the last years, the computational power of mobile devices such as smartphones and tablets has grown dramatically, reaching the level of desktop computers available not long ago. While standard smartphone apps are no longer a problem for them, there is still a group of tasks that can easily challenge even high-end devices, namely running artificial intelligence algorithms. In this paper, we present a study of the current state of deep learning in the Android ecosystem and describe available frameworks, programming models and the limitations of running AI on smartphones. We give an overview of the hardware acceleration resources available on four main mobile chipset platforms: Qualcomm, HiSilicon, MediaTek and Samsung. Additionally, we present the real-world performance results of different mobile SoCs collected with AI Benchmark (http://ai-benchmark.com) that are covering all main existing hardware configurations.

/pdf/ai-benchmark-running-deep-neural-networks-on-android-2x4dsd2dao.pdf

AI Benchmark: Running Deep Neural Networks on Android Smartphones

We introduce a data-free quantization method for deep neural networks that does not require fine-tuning or hyperparameter selection. It achieves near-original model performance on common computer vision architectures and tasks. 8-bit fixed-point quantization is essential for efficient inference on modern deep learning hardware. However, quantizing models to run in 8-bit is a non-trivial task, frequently leading to either significant performance reduction or engineering time spent on training a network to be amenable to quantization. Our approach relies on equalizing the weight ranges in the network by making use of a scale-equivariance property of activation functions. In addition the method corrects biases in the error that are introduced during quantization. This improves quantization accuracy performance, and can be applied to many common computer vision architectures with a straight forward API call. For common architectures, such as the MobileNet family, we achieve state-of-the-art quantized model performance. We further show that the method also extends to other computer vision architectures and tasks such as semantic segmentation and object detection.

/pdf/data-free-quantization-through-weight-equalization-and-bias-dd265ik5kb.pdf

Data-Free Quantization Through Weight Equalization and Bias Correction

Joint image filters can leverage the guidance image as a prior and transfer the structural details from the guidance image to the target image for suppressing noise or enhancing spatial resolution. Existing methods rely on various kinds of explicit filter construction or hand-designed objective functions. It is thus difficult to understand, improve, and accelerate them in a coherent framework. In this paper, we propose a learning-based approach to construct a joint filter based on Convolutional Neural Networks. In contrast to existing methods that consider only the guidance image, our method can selectively transfer salient structures that are consistent in both guidance and target images. We show that the model trained on a certain type of data, e.g., RGB and depth images, generalizes well for other modalities, e.g., Flash/Non-Flash and RGB/NIR images. We validate the effectiveness of the proposed joint filter through extensive comparisons with state-of-the-art methods.

Deep Joint Image Filtering

Pruning and quantization for deep neural network acceleration: A survey

Color, infrared, and flash images captured in different fields can be employed to effectively eliminate noise and other visual artifacts. We propose a two-image restoration framework considering input images in different fields, for example, one noisy color image and one dark-flashed near infrared image. The major issue in such a framework is to handle structure divergence and find commonly usable edges and smooth transition for visually compelling image reconstruction. We introduce a scale map as a competent representation to explicitly model derivative-level confidence and propose new functions and a numerical solver to effectively infer it following new structural observations. Our method is general and shows a principled way for cross-field restoration.

/pdf/cross-field-joint-image-restoration-via-scale-map-3gp4aie1i0.pdf

Cross-Field Joint Image Restoration via Scale Map

Near-infrared (NIR) light has stronger penetration capability than visible light due to its long wavelengths and is thus less scattered by particles in the air This makes it desirable for image dehazing to unveil details of distant objects in landscape photographs In this paper, we propose an improved image dehazing scheme using a pair of color and NIR images, which effectively estimates the airlight color and transfers details from the NIR A two-stage dehazing method is proposed by exploiting the dissimilarity between RGB and NIR for airlight color estimation, followed by a dehazing procedure through an optimization framework Experiments on captured haze images show that our method can achieve substantial improvements on the detail recovery and the color distribution over the existing image dehazing algorithms

/pdf/near-infrared-guided-color-image-dehazing-4eyvt9g7gd.pdf

Near-infrared guided color image dehazing

As deep learning (DL) is being rapidly pushed to edge computing, researchers invented various ways to make inference computation more efficient on mobile/IoT devices, such as network pruning, parameter compression, and etc. Quantization, as one of the key approaches, can effectively offload GPU, and make it possible to deploy DL on fixed-point pipeline. Unfortunately, not all existing networks design are friendly to quantization. For example, the popular lightweight MobileNetV1, while it successfully reduces parameter size and computation latency with separable convolution, our experiment shows its quantized models have large performance gap against its float point models. To resolve this, we analyzed the root cause of quantization loss and proposed a quantization-friendly separable convolution architecture. By evaluating the image classification task on ImageNet2012 dataset, our modified MobileNetV1 model can archive 8-bit inference top-1 accuracy in 68.03%, almost closed the gap to the float pipeline.

A Quantization-Friendly Separable Convolution for MobileNets

Certain aspects relate to systems and techniques for generating high resolution iris templates and for detecting spoofs, enabling more reliable and secure iris authentication. Pairs of RGB and NIR images can be captured by the iris authentication system for use in iris authentication, for example using an NIR LED flash and a four-channel image sensor. Multiple images of the user's iris can be captured by the system in a relatively short period of time and can be fused together to generate a high resolution iris image that can contain more detail of the iris structure and unique pattern than each individual images. The “liveness” of the iris, referring to whether the iris is a real human iris or an iris imitation, can be assessed via a liveness ratio based on comparison of known iris and sclera reflectance properties at various wavelengths to determined sensor responses at those same wavelengths.

Multispectral eye analysis for identity authentication

As deep learning (DL) is being rapidly pushed to edge computing, researchers invented various ways to make inference computation more efficient on mobile/IoT devices, such as network pruning, parameter compression, and etc. Quantization, as one of the key approaches, can effectively offload GPU, and make it possible to deploy DL on fixed-point pipeline. Unfortunately, not all existing networks design are friendly to quantization. For example, the popular lightweight MobileNetV1, while it successfully reduces parameter size and computation latency with separable convolution, our experiment shows its quantized models have large accuracy gap against its float point models. To resolve this, we analyzed the root cause of quantization loss and proposed a quantization-friendly separable convolution architecture. By evaluating the image classification task on ImageNet2012 dataset, our modified MobileNetV1 model can archive 8-bit inference top-1 accuracy in 68.03%, almost closed the gap to the float pipeline.

Liang Shen

Papers

Cross-Field Joint Image Restoration via Scale Map

Near-infrared guided color image dehazing

A Quantization-Friendly Separable Convolution for MobileNets

Multispectral eye analysis for identity authentication

A Quantization-Friendly Separable Convolution for MobileNets