DynaQuant: Dynamic Mixed-Precision Quantization for Learned Image Compression

Recently, learned image compression methods have shown their outstanding rate-distortion performance when compared to traditional frameworks. Although numerous progress has been made in learned image compression, the computation cost is still at a high level. To address this problem, we propose AdderIC, which utilizes adder neural networks (AdderNet) to construct an image compression framework. According to the characteristics of image compression, we introduce several strategies to improve the performance of AdderNet in this field. Specifically, Haar Wavelet Transform is adopted to make AdderIC learn high-frequency information efficiently. In addition, implicit deconvolution with the kernel size of 1 is applied after each adder layer to reduce spatial redundancies. Moreover, we develop a novel Adder-ID-PixelShuffle cascade upsampling structure to remove checkerboard artifacts. Experiments demonstrate that our AdderIC model can largely outperform conventional AdderNet when applied in image compression and achieve comparable rate-distortion performance to that of its CNN baseline with about 80% multiplication FLOPs and 30% energy consumption reduction.

This paper investigates the problem of efficient compression and high-fidelity reconstruction of three-dimensional (3D) microscopic measurement data, which typically consist of massive high-resolution images. Conventional image compression standards (e.g., JPEG, PNG, VVC) struggle to preserve the fine microstructural details essential for accurate measurement and analysis. To address this limitation, we explore learned image compression (LIC) as a domain-adaptive solution for the compression–transmission–decompression pipeline of 3D microscopic imaging. We further propose a novel channel-wise entropy regularization strategy that enforces neighboring pixels within the same feature channel to share consistent representations, thereby exploiting the intrinsic local homogeneity of microscopic textures. By integrating convolutional neural networks for local feature extraction with Transformer-based attention mechanisms for global dependency modeling, the proposed method achieves both efficient entropy modeling and improved reconstruction accuracy. Experimental results demonstrate superior rate–distortion performance compared with conventional schemes, while significantly enhancing the continuity and consistency of reconstructed microstructures. These findings highlight the potential of LIC with domain-specific regularization to enable next-generation high-precision microscopic measurement systems.

Learned image compression (LIC) techniques have achieved remarkable progress; however, effectively integrating high-level semantic information remains challenging. In this work, we present a Semantic-Enhanced Learned Image Compression framework, termed SELIC, which leverages high-level textual guidance to improve rate-distortion performance. Specifically, SELIC employs a text encoder to extract rich semantic descriptions from the input image. These textual features are transformed into fixed-dimension tensors and seamlessly fused with the image-derived latent representation. By embedding the SELIC tensor directly into the compression pipeline, our approach enriches the bitstream without requiring additional inputs at the decoder, thereby maintaining fast and efficient decoding. Extensive experiments on benchmark datasets (e.g., Kodak) demonstrate that integrating semantic information substantially enhances compression quality. Our SELIC-guided method outperforms a baseline LIC model without semantic integration by approximately 0.1-0.15 dB across a wide range of bit rates in PSNR and achieves a 4.9% BD-rate improvement over VVC. Moreover, this improvement comes with minimal computational overhead, making the proposed SELIC framework a practical solution for advanced image compression applications.

Rate control (RC) is a critical component in learned image compression (LIC), particularly in the emerging JPEG-AI standard, which enables adaptive bitrate achievement to meet diverse bandwidth constraints. JPEG-AI default RC employs an iterative optimization process, wherein a pre-trained RC model is selected and the (generated) latent representations are adjusted based on the mismatch between actual and target bitrates. Despite satisfactory results, such a trial-and-error paradigm necessitates multiple processing cycles, resulting in inevitable computational overhead. We propose an efficient neural rate control framework for JPEG-AI to address this limitation. Our idea is to train a ResNet-based neural control (NRC) to learn the mapping from the input images and target bitrates to the optimal coding parameters. The trained NRC can then be applied to predict the coding parameters based on the new input images and target bitrates directly. Experimental results on DIV2K and MSCOCO datasets show that our NRC achieves comparable rate-distortion performance while reducing encoding time by about 5× compared to JPEG-AI default RC.

There have been many compression standards developed during the past few decades and technological advances has resulted in introducing many methodologies with promising results. As far as PSNR metric is concerned, there is a performance gap between reigning compression standards and learned compression algorithms. Based on research, we experimented using an accurate entropy model on the learned compression algorithms to determine the rate-distortion performance. In this paper, discretized Gaussian Mixture likelihood is proposed to determine the latent code parameters in order to attain a more flexible and accurate model of entropy. Moreover, we have also enhanced the performance of the work by introducing recent attention modules in the network architecture. Simulation results indicate that when compared with the previously existing techniques using high-resolution and Kodak datasets, the proposed work achieves a higher rate of performance. When MS-SSIM is used for optimization, our work generates a more visually pleasant image.

Reduction operations for constraint satisfaction

In this paper, five mathematical models of single neurons are discussed and compared. The physical meanings, derivations, and differential equations of each model are provided. Since for many applications the spiking rates of neurons are of great importance, we compare the spiking rate patterns under different sustained current inputs. Numerical stability and accuracy are also considered. The computational cost and storage requirements needed to numerically solve each of the models are also discussed.

Image compression based on transform coding appears to be approaching an asymptotic bit rate limit for application-specific distortion levels. However, a new compression technology, called object-based compression (OBC) promises improved rate-distortion performance at higher compression ratios. OBC involves segmentation of image regions, followed by efficient encoding of each region’s content and boundary. Advantages of OBC include efficient representation of commonly occurring textures and shapes in terms of pointers into a compact codebook of region contents and boundary primitives. This facilitates fast decompression via substitution, at the cost of codebook search in the compression step. Segmentation cose and error are significant disadvantages in current OBC implementations. Several innovative techniques have been developed for region segmentation, including (a) moment-based analysis, (b) texture representation in terms of a syntactic grammar, and (c) transform coding approaches such as wavelet based compression used in MPEG-7 or JPEG-2000. Region-based characterization with variance templates is better understood, but lacks the locality of wavelet representations. In practice, tradeoffs are made between representational fidelity, computational cost, and storage requirement. This paper overviews current techniques for automatic region segmentation and representation, especially those that employ wavelet classification and region growing techniques. Implementational discussion focuses on complexity measures and performance metrics such as segmentation error and computational cost.

This paper presents the optimum bit allocation for fixed rate pyramid coding structures which use a lowpass filter with nominal bandwidth of /spl pi//4 for predecimation filtering. This technique results in about 32% less in the computational complexity and 20% less in the memory requirements than the conventional filters used for pyramid coding. Three different algorithms are introduced. In the first algorithm the decimated image is transmitted at 8 bpp and the difference image is DCT-encoded. In the second algorithm the decimated image is encoded using BTC and the difference image is DCT-encoded, while in the third algorithm the decimated image is BTC-encoded with further quantization of the DCT coefficients of the difference image. These algorithms achieve high compression rate and allow at the same time a low complexity encoder.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

In the multiview video plus depth (MVD) coding format, both texture and depth views are jointly compressed to represent the 3-D video content. The MVD format enables synthesis of virtual views through depth-image-based rendering; hence, distortion in the texture and depth views affects the quality of the synthesized virtual views. Bit allocation between texture and depth views has been studied with some promising results. However, to the best of our knowledge, most of the existing bit-allocation methods attempt to allocate a fixed amount of total bit rate between texture and depth views; that is, to select appropriate pair of quantization parameters for texture and depth views to maximize the synthesized view quality subject to a fixed total bit rate. In this paper we propose a scalable bit-allocation scheme, where a single ordering of texture and depth packets is derived and used to obtain optimal bit allocation between texture and depth views for any total target rates. In the proposed scheme, both texture and depth views are encoded using the quality scalable coding method; that is, medium grain scalable (MGS) coding of the Scalable Video Coding (SVC) extension of the Advanced Video Coding (H.264/AVC) standard. For varying target total bit rates, optimal bit truncation points for both texture and depth views can be obtained using the proposed scheme. Moreover, we propose to order the enhancement layer packets of the H.264/SVC MGS encoded depth view according to their contribution to the reduction of the synthesized view distortion. On one hand, this improves the depth view packet ordering when considered the rate-distortion performance of synthesized views, which is demonstrated by the experimental results. On the other hand, the information obtained in this step is used to facilitate optimal bit allocation between texture and depth views. Experimental results demonstrate the effectiveness of the proposed scalable bit-allocation scheme for texture and depth views.

This paper addresses the problem of efficient coding of an important class of signals, namely piecewise polynomials. For this signal class, we develop a coding algorithm, which achieves oracle like rate-distortion (R-D) behavior in the high bit rate regime and with a reasonable computational complexity. For the 1-D case, our scheme is based on the binary tree segmentation of the signal and an optimal bit allocation strategy among the different signal segments. The scheme further encodes the similar neighbors jointly to achieve the right exponentially decaying R-D behavior (D(R) /spl sim/ c/sub 0/2/sup -c1R/). We have also shown that the computational cost of the scheme is of the order O(N log N). We then show that the scheme can be easily extended to the 2-D case, as the quadtree based coding scheme, with the similar R-D behavior and computational cost. Finally, we conclude with some numerical results.

The modern era of embedded system design is geared towards design of low-power systems. One way to reduce power in an ASIC implementation is to reduce the bit-width precision of its computation units. This paper describes algorithms to optimize the bit-widths of fixed point variables for low power in a SystemC design environment. We propose an algorithm for optimal bit width precision for two variables and a greedy heuristic which works for any number of variables. The algorithms are used in the automation of converting floating point SystemC programs into ASIC synthesizable SystemC programs. Expected inputs are profiled to estimate errors in the finite precision conversions. Experimental results on the trade-offs between quantization error, power consumption and hardware resources used are reported on a set of four SystemC benchmarks that are mapped onto 0.18 micron ASIC cell library from Artisan Components. We demonstrate that it is possible to reduce the power consumption by 50% on average by allowing round-off errors to increase from 0.5% to 1%. 1

Efficient parallel factorization and solution of structured and unstructured linear systems

In this paper, we address the dependent joint bit allocation problem in H.264/AVC statistical multiplexing. For most of the existing methods, in order to improve the overall visual quality, the bit allocation is performed upon the relative complexities of different video programs. However, due to the temporal prediction employed in H.264, the influence of current frame to the rate-distortion (R-D) performances of the future frames should be taken in to account as well. The contributions of the paper are two-fold. First, a simple but accurate inter-frame dependency model (IFDM) is introduced which can quantitatively measure the coding dependency between the current coding frame and its reference frame. Second, based on the IFDM, the dependent joint bit allocation problem is revisited, and both the frame complexity and inter-frame dependency are considered in the bit allocation process. Then, it is proved that the dependent joint bit allocation problem can actually be relaxed into a convex optimization problem which can be optimally and efficiently solved. Experimental results demonstrate that the proposed dependent bit allocation method achieves 33.45% and 11.10% bitrate reduction on average compared with the equivalent bit allocation (EBA) and the optimal independent joint bit allocation (OIJBA) methods respectively.

The primary dilemmas in nonparametric algorithms like k-nearest neighbor classification are the largest computational and storage requirements. Moreover, the effectiveness of classification decreases due to uneven distribution of training data. In this paper, we present three approaches to minimize computation time and storage requirements. In order to achieve the goal, we present three approaches: fast k-NN, training set reduction techniques, and a hybrid of the previous two approaches. We have compared three approaches to existing methods and results show that the effectiveness (in terms of execution time and storage requirement) of the three algorithms are significantly better than existing algorithms.