In this work, we reveal the limitations of visual tokenizers and VAEs in preserving fine-grained features, and propose a benchmark to evaluate reconstruction performance for two challenging visual contents: text and face. Visual tokenizers and VAEs have significantly advanced visual generation and multimodal modeling by providing more efficient compressed or quantized image representations. However, while helping production models reduce computational burdens, the information loss from image compression fundamentally limits the upper bound of visual generation quality. To evaluate this upper bound, we focus on assessing reconstructed text and facial features since they typically: 1) exist at smaller scales, 2) contain dense and rich textures, 3) are prone to collapse, and 4) are highly sensitive to human vision. We first collect and curate a diverse set of clear text and face images from existing datasets. Unlike approaches using VLM models, we employ established OCR and face recognition models for evaluation, ensuring accuracy while maintaining an exceptionally lightweight assessment process <span style="font-weight: bold; color: rgb(214, 21, 21);">requiring just 2GB memory and 4 minutes</span> to complete. Using our benchmark, we analyze text and face reconstruction quality across various scales for different image tokenizers and VAEs. Our results show modern visual tokenizers still struggle to preserve fine-grained features, especially at smaller scales. We further extend this evaluation framework to video, conducting comprehensive analysis of video tokenizers. Additionally, we demonstrate that traditional metrics fail to accurately reflect reconstruction performance for faces and text, while our proposed metrics serve as an effective complement.
While recent diffusion-based generative image codecs have shown impressive performance, their iterative sampling process introduces unpleasing latency. In this work, we revisit the design of a diffusion-based codec and argue that multi-step sampling is not necessary for generative compression. Based on this insight, we propose OneDC, a One-step Diffusion-based generative image Codec -- that integrates a latent compression module with a one-step diffusion generator. Recognizing the critical role of semantic guidance in one-step diffusion, we propose using the hyperprior as a semantic signal, overcoming the limitations of text prompts in representing complex visual content. To further enhance the semantic capability of the hyperprior, we introduce a semantic distillation mechanism that transfers knowledge from a pretrained generative tokenizer to the hyperprior codec. Additionally, we adopt a hybrid pixel- and latent-domain optimization to jointly enhance both reconstruction fidelity and perceptual realism. Extensive experiments demonstrate that OneDC achieves SOTA perceptual quality even with one-step generation, offering over 40% bitrate reduction and 20x faster decoding compared to prior multi-step diffusion-based codecs. Code will be released later.
Most existing approaches for image and video compression perform transform coding in the pixel space to reduce redundancy. However, due to the misalignment between the pixel-space distortion and human perception, such schemes often face the difficulties in achieving both high-realism and high-fidelity at ultra-low bitrate. To solve this problem, we propose \textbf{G}enerative \textbf{L}atent \textbf{C}oding (\textbf{GLC}) models for image and video compression, termed GLC-image and GLC-Video. The transform coding of GLC is conducted in the latent space of a generative vector-quantized variational auto-encoder (VQ-VAE). Compared to the pixel-space, such a latent space offers greater sparsity, richer semantics and better alignment with human perception, and show its advantages in achieving high-realism and high-fidelity compression. To further enhance performance, we improve the hyper prior by introducing a spatial categorical hyper module in GLC-image and a spatio-temporal categorical hyper module in GLC-video. Additionally, the code-prediction-based loss function is proposed to enhance the semantic consistency. Experiments demonstrate that our scheme shows high visual quality at ultra-low bitrate for both image and video compression. For image compression, GLC-image achieves an impressive bitrate of less than $0.04$ bpp, achieving the same FID as previous SOTA model MS-ILLM while using $45\%$ fewer bitrate on the CLIC 2020 test set. For video compression, GLC-video achieves 65.3\% bitrate saving over PLVC in terms of DISTS.
Pretrained latent diffusion models have shown strong potential for lossy image compression, owing to their powerful generative priors. Most existing diffusion-based methods reconstruct images by iteratively denoising from random noise, guided by compressed latent representations. While these approaches have achieved high reconstruction quality, their multi-step sampling process incurs substantial computational overhead. Moreover, they typically require training separate models for different compression bit-rates, leading to significant training and storage costs. To address these challenges, we propose a one-step diffusion codec across multiple bit-rates. termed OSCAR. Specifically, our method views compressed latents as noisy variants of the original latents, where the level of distortion depends on the bit-rate. This perspective allows them to be modeled as intermediate states along a diffusion trajectory. By establishing a mapping from the compression bit-rate to a pseudo diffusion timestep, we condition a single generative model to support reconstructions at multiple bit-rates. Meanwhile, we argue that the compressed latents retain rich structural information, thereby making one-step denoising feasible. Thus, OSCAR replaces iterative sampling with a single denoising pass, significantly improving inference efficiency. Extensive experiments demonstrate that OSCAR achieves superior performance in both quantitative and visual quality metrics. The code and models will be released at https://github.com/jp-guo/OSCAR.
The exponential growth of visual data in digital communications has intensified the need for efficient compression techniques that balance rate-distortion performance with computational feasibility. While recent neural compression approaches have shown promise, they still struggle with fundamental challenges: preserving perceptual quality at high compression ratios, computational efficiency, and adaptability to diverse visual content. This paper introduces GANCompress, a novel neural compression framework that synergistically combines Binary Spherical Quantization (BSQ) with Generative Adversarial Networks (GANs) to address these challenges. Our approach employs a transformer-based autoencoder with an enhanced BSQ bottleneck that projects latent representations onto a hypersphere, enabling efficient discretization with bounded quantization error. This is followed by a specialized GAN architecture incorporating frequency-domain attention and color consistency optimization. Experimental results demonstrate that GANCompress achieves substantial improvement in compression efficiency -- reducing file sizes by up to 100x with minimal visual distortion. Our method outperforms traditional codecs like H.264 by 12-15% in perceptual metrics while maintaining comparable PSNR/SSIM values, with 2.4x faster encoding and decoding speeds. On standard benchmarks including ImageNet-1k and COCO2017, GANCompress sets a new state-of-the-art, reducing FID from 0.72 to 0.41 (43% improvement) compared to previous methods while maintaining higher throughput. This work presents a significant advancement in neural compression technology with promising applications for real-time visual communication systems.
Denoising diffusion models achieved impressive results on several image generation tasks often outperforming GAN based models. Recently, the generative capabilities of diffusion models have been employed for perceptual image compression, such as in CDC. A major drawback of these diffusion-based methods is that, while producing impressive perceptual quality images they are dropping in fidelity/increasing the distortion to the original uncompressed images when compared with other traditional or learned image compression schemes aiming for fidelity. In this paper, we propose a hybrid compression scheme optimized for perceptual quality, extending the approach of the CDC model with a decoder network in order to reduce the impact on distortion metrics such as PSNR. After using the decoder network to generate an initial image, optimized for distortion, the latent conditioned diffusion model refines the reconstruction for perceptual quality by predicting the residual. On standard benchmarks, we achieve up to +2dB PSNR fidelity improvements while maintaining comparable LPIPS and FID perceptual scores when compared with CDC. Additionally, the approach is easily extensible to video compression, where we achieve similar results.
With the increasing exploration and exploitation of the underwater world, underwater images have become a critical medium for human interaction with marine environments, driving extensive research into their efficient transmission and storage. However, contemporary underwater image compression algorithms fail to fully leverage the unique characteristics distinguishing underwater scenes from terrestrial images, resulting in suboptimal performance. To address this limitation, we introduce HQUIC, designed to exploit underwater-image-specific features for enhanced compression efficiency. HQUIC employs an ALTC module to adaptively predict the attenuation coefficients and global light information of the images, which effectively mitigates the issues caused by the differences in lighting and tone existing in underwater images. Subsequently, HQUIC employs a codebook as an auxiliary branch to extract the common objects within underwater images and enhances the performance of the main branch. Furthermore, HQUIC dynamically weights multi-scale frequency components, prioritizing information critical for distortion quality while discarding redundant details. Extensive evaluations on diverse underwater datasets demonstrate that HQUIC outperforms state-of-the-art compression methods.
Learned image compression methods have attracted great research interest and exhibited superior rate-distortion performance to the best classical image compression standards of the present. The entropy model plays a key role in learned image compression, which estimates the probability distribution of the latent representation for further entropy coding. Most existing methods employed hyper-prior and auto-regressive architectures to form their entropy models. However, they only aimed to explore the internal dependencies of latent representation while neglecting the importance of extracting prior from training data. In this work, we propose a novel entropy model named Dictionary-based Cross Attention Entropy model, which introduces a learnable dictionary to summarize the typical structures occurring in the training dataset to enhance the entropy model. Extensive experimental results have demonstrated that the proposed model strikes a better balance between performance and latency, achieving state-of-the-art results on various benchmark datasets.
Existing multimodal large model-based image compression frameworks often rely on a fragmented integration of semantic retrieval, latent compression, and generative models, resulting in suboptimal performance in both reconstruction fidelity and coding efficiency. To address these challenges, we propose a residual-guided ultra lowrate image compression named ResULIC, which incorporates residual signals into both semantic retrieval and the diffusion-based generation process. Specifically, we introduce Semantic Residual Coding (SRC) to capture the semantic disparity between the original image and its compressed latent representation. A perceptual fidelity optimizer is further applied for superior reconstruction quality. Additionally, we present the Compression-aware Diffusion Model (CDM), which establishes an optimal alignment between bitrates and diffusion time steps, improving compression-reconstruction synergy. Extensive experiments demonstrate the effectiveness of ResULIC, achieving superior objective and subjective performance compared to state-of-the-art diffusion-based methods with - 80.7%, -66.3% BD-rate saving in terms of LPIPS and FID. Project page is available at https: //njuvision.github.io/ResULIC/.
Recently, perceptual image compression has achieved significant advancements, delivering high visual quality at low bitrates for natural images. However, for screen content, existing methods often produce noticeable artifacts when compressing text. To tackle this challenge, we propose versatile perceptual screen image compression with diffusion rendering (PICD), a codec that works well for both screen and natural images. More specifically, we propose a compression framework that encodes the text and image separately, and renders them into one image using diffusion model. For this diffusion rendering, we integrate conditional information into diffusion models at three distinct levels: 1). Domain level: We fine-tune the base diffusion model using text content prompts with screen content. 2). Adaptor level: We develop an efficient adaptor to control the diffusion model using compressed image and text as input. 3). Instance level: We apply instance-wise guidance to further enhance the decoding process. Empirically, our PICD surpasses existing perceptual codecs in terms of both text accuracy and perceptual quality. Additionally, without text conditions, our approach serves effectively as a perceptual codec for natural images.
With the widespread application of facial image data across various domains, the efficient storage and transmission of facial images has garnered significant attention. However, the existing learned face image compression methods often produce unsatisfactory reconstructed image quality at low bit rates. Simply adapting diffusion-based compression methods to facial compression tasks results in reconstructed images that perform poorly in downstream applications due to insufficient preservation of high-frequency information. To further explore the diffusion prior in facial image compression, we propose Facial Image Compression with a Stable Diffusion Prior (FaSDiff), a method that preserves consistency through frequency enhancement. FaSDiff employs a high-frequency-sensitive compressor in an end-to-end framework to capture fine image details and produce robust visual prompts. Additionally, we introduce a hybrid low-frequency enhancement module that disentangles low-frequency facial semantics and stably modulates the diffusion prior alongside visual prompts. The proposed modules allow FaSDiff to leverage diffusion priors for superior human visual perception while minimizing performance loss in machine vision due to semantic inconsistency. Extensive experiments show that FaSDiff outperforms state-of-the-art methods in balancing human visual quality and machine vision accuracy. The code will be released after the paper is accepted.
The rapid development of AIGC foundation models has revolutionized the paradigm of image compression, which paves the way for the abandonment of most pixel-level transform and coding, compelling us to ask: why compress what you can generate if the AIGC foundation model is powerful enough to faithfully generate intricate structure and fine-grained details from nothing more than some compact descriptors, i.e., texts, or cues. Fortunately, recent GPT-4o image generation of OpenAI has achieved impressive cross-modality generation, editing, and design capabilities, which motivates us to answer the above question by exploring its potential in image compression fields. In this work, we investigate two typical compression paradigms: textual coding and multimodal coding (i.e., text + extremely low-resolution image), where all/most pixel-level information is generated instead of compressing via the advanced GPT-4o image generation function. The essential challenge lies in how to maintain semantic and structure consistency during the decoding process. To overcome this, we propose a structure raster-scan prompt engineering mechanism to transform the image into textual space, which is compressed as the condition of GPT-4o image generation. Extensive experiments have shown that the combination of our designed structural raster-scan prompts and GPT-4o's image generation function achieved the impressive performance compared with recent multimodal/generative image compression at ultra-low bitrate, further indicating the potential of AIGC generation in image compression fields.
Current learned image compression models typically exhibit high complexity, which demands significant computational resources. To overcome these challenges, we propose an innovative approach that employs hierarchical feature extraction transforms to significantly reduce complexity while preserving bit rate reduction efficiency. Our novel architecture achieves this by using fewer channels for high spatial resolution inputs/feature maps. On the other hand, feature maps with a large number of channels have reduced spatial dimensions, thereby cutting down on computational load without sacrificing performance. This strategy effectively reduces the forward pass complexity from \(1256 \, \text{kMAC/Pixel}\) to just \(270 \, \text{kMAC/Pixel}\). As a result, the reduced complexity model can open the way for learned image compression models to operate efficiently across various devices and pave the way for the development of new architectures in image compression technology.
Recent advancements in learned image compression (LIC) have yielded impressive performance gains. Notably, the learned image compression models with multi-reference entropy models (MLIC series) have significantly outperformed existing traditional image codecs such as the Versatile Video Coding (VVC) Intra. In this paper, we present MLICv2 and MLICv2$^+$, enhanced versions of the MLIC series, featuring improved transform techniques, entropy modeling, and instance adaptability. For better transform, we introduce a simple token mixing transform block inspired by the meta transformer architecture, addressing the performance degradation at high bit-rates observed in previous MLIC series while maintaining computational efficiency. To enhance entropy modeling, we propose a hyperprior-guided global correlation prediction, enabling the capture of global contexts in the initial slice of the latent representation. We also develop a channel reweighting module to dynamically prioritize important channels within each context. Additionally, advanced positional embedding for context modeling and selective compression with guided optimization are investigated. To boost instance adaptability, we employ stochastic Gumbel annealing to iteratively refine the latent representation according to the rate-distortion optimization of a specific input image. This approach further enhances performance without impacting decoding speed. Experimental results demonstrate that our MLICv2 and MLICv2$^+$ achieve state-of-the-art performance, reducing Bjontegaard-Delta rate (BD-rate) by 16.54%, 21.61%, 16.05% and 20.46%, 24.35%, 19.14% respectively, compared to VTM-17.0 Intra on the Kodak, Tecnick, CLIC Pro Val dataset, respectively.
Tokenized visual representations have shown great promise in image compression, yet their extension to video remains underexplored due to the challenges posed by complex temporal dynamics and stringent bitrate constraints. In this paper, we propose Tokenized Video Compression (TVC), the first token-based dual-stream video compression framework designed to operate effectively at ultra-low bitrates. TVC leverages the powerful Cosmos video tokenizer to extract both discrete and continuous token streams. The discrete tokens (i.e., code maps generated by FSQ) are partially masked using a strategic masking scheme, then compressed losslessly with a discrete checkerboard context model to reduce transmission overhead. The masked tokens are reconstructed by a decoder-only transformer with spatiotemporal token prediction. Meanwhile, the continuous tokens, produced via an autoencoder (AE), are quantized and compressed using a continuous checkerboard context model, providing complementary continuous information at ultra-low bitrate. At the Decoder side, both streams are fused using ControlNet, with multi-scale hierarchical integration to ensure high perceptual quality alongside strong fidelity in reconstruction. This work mitigates the long-standing skepticism about the practicality of tokenized video compression and opens up new avenues for semantics-aware, token-native video compression.
Neural image compression, necessary in various machine-to-machine communication scenarios, suffers from its heavy encode-decode structures and inflexibility in switching between different compression levels. Consequently, it raises significant challenges in applying the neural image compression to edge devices that are developed for powerful servers with high computational and storage capacities. We take a step to solve the challenges by proposing a new transformer-based edge-compute-free image coding framework called Easz. Easz shifts the computational overhead to the server, and hence avoids the heavy encoding and model switching overhead on the edge. Easz utilizes a patch-erase algorithm to selectively remove image contents using a conditional uniform-based sampler. The erased pixels are reconstructed on the receiver side through a transformer-based framework. To further reduce the computational overhead on the receiver, we then introduce a lightweight transformer-based reconstruction structure to reduce the reconstruction load on the receiver side. Extensive evaluations conducted on a real-world testbed demonstrate multiple advantages of Easz over existing compression approaches, in terms of adaptability to different compression levels, computational efficiency, and image reconstruction quality.
Neural compression methods are gaining popularity due to their superior rate-distortion performance over traditional methods, even at extremely low bitrates below 0.1 bpp. As deep learning architectures, these models are prone to bias during the training process, potentially leading to unfair outcomes for individuals in different groups. In this paper, we present a general, structured, scalable framework for evaluating bias in neural image compression models. Using this framework, we investigate racial bias in neural compression algorithms by analyzing nine popular models and their variants. Through this investigation, we first demonstrate that traditional distortion metrics are ineffective in capturing bias in neural compression models. Next, we highlight that racial bias is present in all neural compression models and can be captured by examining facial phenotype degradation in image reconstructions. We then examine the relationship between bias and realism in the decoded images and demonstrate a trade-off across models. Finally, we show that utilizing a racially balanced training set can reduce bias but is not a sufficient bias mitigation strategy. We additionally show the bias can be attributed to compression model bias and classification model bias. We believe that this work is a first step towards evaluating and eliminating bias in neural image compression models.
IoT devices have limited hardware capabilities and are often deployed in remote areas. Consequently, advanced vision models surpass such devices' processing and storage capabilities, requiring offloading of such tasks to the cloud. However, remote areas often rely on LPWANs technology with limited bandwidth, high packet loss rates, and extremely low duty cycles, which makes fast offloading for time-sensitive inference challenging. Today's approaches, which are deployable on weak devices, generate a non-progressive bit stream, and therefore, their decoding quality suffers strongly when data is only partially available on the cloud at a deadline due to limited bandwidth or packet losses. In this paper, we introduce LimitNet, a progressive, content-aware image compression model designed for extremely weak devices and networks. LimitNet's lightweight progressive encoder prioritizes critical data during transmission based on the content of the image, which gives the cloud the opportunity to run inference even with partial data availability. Experimental results demonstrate that LimitNet, on average, compared to SOTA, achieves 14.01 p.p. (percentage point) higher accuracy on ImageNet1000, 18.01 pp on CIFAR100, and 0.1 higher [email omitted] on COCO. Also, on average, LimitNet saves 61.24% bandwidth on ImageNet1000, 83.68% on CIFAR100, and 42.25% on the COCO dataset compared to SOTA, while it only has 4% more encoding time compared to JPEG (with a fixed quality) on STM32F7 (Cortex-M7).
In lossy compression, the classical tradeoff between compression rate and reconstruction distortion has traditionally guided algorithm design. However, Blau and Michaeli [5] introduced a generalized framework, known as the rate-distortion-perception (RDP) function, incorporating perceptual quality as an additional dimension of evaluation. More recently, the rate-distortion-classification (RDC) function was investigated in [19], evaluating compression performance by considering classification accuracy alongside distortion. In this paper, we explore universal representations, where a single encoder is developed to achieve multiple decoding objectives across various distortion and classification (or perception) constraints. This universality avoids retraining encoders for each specific operating point within these tradeoffs. Our experimental validation on the MNIST dataset indicates that a universal encoder incurs only minimal performance degradation compared to individually optimized encoders for perceptual image compression tasks, aligning with prior results from [23]. Nonetheless, we also identify that in the RDC setting, reusing an encoder optimized for one specific classification-distortion tradeoff leads to a significant distortion penalty when applied to alternative points.
Efficiently transferring Learned Image Compression (LIC) model from human perception to machine perception is an emerging challenge in vision-centric representation learning. Existing approaches typically adapt LIC to downstream tasks in a single-task manner, which is inefficient, lacks task interaction, and results in multiple task-specific bitstreams. To address these limitations, we propose an asymmetric adaptor framework that supports multi-task adaptation within a single model. Our method introduces a shared adaptor to learn general semantic features and task-specific adaptors to preserve task-level distinctions. With only lightweight plug-in modules and a frozen base codec, our method achieves strong performance across multiple tasks while maintaining compression efficiency. Experiments on the PASCAL-Context benchmark demonstrate that our method outperforms both Fully Fine-Tuned and other Parameter Efficient Fine-Tuned (PEFT) baselines, and validating the effectiveness of multi-vision transferring.