Miao Cao

Lishun Wang, Miao Cao, Yong Zhong et al.

Video snapshot compressive imaging (SCI) captures multiple sequential video frames by a single measurement using the idea of computational imaging. The underlying principle is to modulate high-speed frames through different masks and these modulated frames are summed to a single measurement captured by a low-speed 2D sensor (dubbed optical encoder); following this, algorithms are employed to reconstruct the desired high-speed frames (dubbed software decoder) if needed. In this paper, we consider the reconstruction algorithm in video SCI, i.e., recovering a series of video frames from a compressed measurement. Specifically, we propose a Spatial-Temporal transFormer (STFormer) to exploit the correlation in both spatial and temporal domains. STFormer network is composed of a token generation block, a video reconstruction block, and these two blocks are connected by a series of STFormer blocks. Each STFormer block consists of a spatial self-attention branch, a temporal self-attention branch and the outputs of these two branches are integrated by a fusion network. Extensive results on both simulated and real data demonstrate the state-of-the-art performance of STFormer. The code and models are publicly available at https://github.com/ucaswangls/STFormer.git

Miao Cao, Lishun Wang, Huan Wang et al.

Video Snapshot Compressive Imaging (SCI) aims to use a low-speed 2D camera to capture high-speed scene as snapshot compressed measurements, followed by a reconstruction algorithm to reconstruct the high-speed video frames. State-of-the-art (SOTA) deep learning-based algorithms have achieved impressive performance, yet with heavy computational workload. Network quantization is a promising way to reduce computational cost. However, a direct low-bit quantization will bring large performance drop. To address this challenge, in this paper, we propose a simple low-bit quantization framework (dubbed Q-SCI) for the end-to-end deep learning-based video SCI reconstruction methods which usually consist of a feature extraction, feature enhancement, and video reconstruction module. Specifically, we first design a high-quality feature extraction module and a precise video reconstruction module to extract and propagate high-quality features in the low-bit quantized model. In addition, to alleviate the information distortion of the Transformer branch in the quantized feature enhancement module, we introduce a shift operation on the query and key distributions to further bridge the performance gap. Comprehensive experimental results manifest that our Q-SCI framework can achieve superior performance, e.g., 4-bit quantized EfficientSCI-S derived by our Q-SCI framework can theoretically accelerate the real-valued EfficientSCI-S by 7.8X with only 2.3% performance gap on the simulation testing datasets. Code is available at https://github.com/mcao92/QuantizedSCI.

Miao Cao, Lishun Wang, Huan Wang et al.

Video Snapshot Compressive Imaging (SCI) uses a low-speed 2D camera to capture high-speed scenes as snapshot compressed measurements, followed by a reconstruction algorithm to retrieve the high-speed video frames. The fast evolving mobile devices and existing high-performance video SCI reconstruction algorithms motivate us to develop mobile reconstruction methods for real-world applications. Yet, it is still challenging to deploy previous reconstruction algorithms on mobile devices due to the complex inference process, let alone real-time mobile reconstruction. To the best of our knowledge, there is no video SCI reconstruction model designed to run on the mobile devices. Towards this end, in this paper, we present an effective approach for video SCI reconstruction, dubbed MobileSCI, which can run at real-time speed on the mobile devices for the first time. Specifically, we first build a U-shaped 2D convolution-based architecture, which is much more efficient and mobile-friendly than previous state-of-the-art reconstruction methods. Besides, an efficient feature mixing block, based on the channel splitting and shuffling mechanisms, is introduced as a novel bottleneck block of our proposed MobileSCI to alleviate the computational burden. Finally, a customized knowledge distillation strategy is utilized to further improve the reconstruction quality. Extensive results on both simulated and real data show that our proposed MobileSCI can achieve superior reconstruction quality with high efficiency on the mobile devices. Particularly, we can reconstruct a 256 X 256 X 8 snapshot compressed measurement with real-time performance (about 35 FPS) on an iPhone 15. Code is available at https://github.com/mcao92/MobileSCI.

Daoyu Li, Hanwen Xu, Miao Cao et al.

Computational reconstruction plays a vital role in computer vision and computational photography. Most of the conventional optimization and deep learning techniques explore local information for reconstruction. Recently, nonlocal low-rank (NLR) reconstruction has achieved remarkable success in improving accuracy and generalization. However, the computational cost has inhibited NLR from seeking global structural similarity, which consequentially keeps it trapped in the tradeoff between accuracy and efficiency and prevents it from high-dimensional large-scale tasks. To address this challenge, we report here the global low-rank (GLR) optimization technique, realizing highly-efficient large-scale reconstruction with global self-similarity. Inspired by the self-attention mechanism in deep learning, GLR extracts exemplar image patches by feature detection instead of conventional uniform selection. This directly produces key patches using structural features to avoid burdensome computational redundancy. Further, it performs patch matching across the entire image via neural-based convolution, which produces the global similarity heat map in parallel, rather than conventional sequential block-wise matching. As such, GLR improves patch grouping efficiency by more than one order of magnitude. We experimentally demonstrate GLR's effectiveness on temporal, frequency, and spectral dimensions, including different computational imaging modalities of compressive temporal imaging, magnetic resonance imaging, and multispectral filter array demosaicing. This work presents the superiority of inherent fusion of deep learning strategies and iterative optimization, and breaks the persistent dilemma of the tradeoff between accuracy and efficiency for various large-scale reconstruction tasks.

Meituan LongCat Team, Bairui Wang, Bayan et al.

We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.

Lishun Wang, Miao Cao, Xin Yuan

Video snapshot compressive imaging (SCI) uses a two-dimensional detector to capture consecutive video frames during a single exposure time. Following this, an efficient reconstruction algorithm needs to be designed to reconstruct the desired video frames. Although recent deep learning-based state-of-the-art (SOTA) reconstruction algorithms have achieved good results in most tasks, they still face the following challenges due to excessive model complexity and GPU memory limitations: 1) these models need high computational cost, and 2) they are usually unable to reconstruct large-scale video frames at high compression ratios. To address these issues, we develop an efficient network for video SCI by using dense connections and space-time factorization mechanism within a single residual block, dubbed EfficientSCI. The EfficientSCI network can well establish spatial-temporal correlation by using convolution in the spatial domain and Transformer in the temporal domain, respectively. We are the first time to show that an UHD color video with high compression ratio can be reconstructed from a snapshot 2D measurement using a single end-to-end deep learning model with PSNR above 32 dB. Extensive results on both simulation and real data show that our method significantly outperforms all previous SOTA algorithms with better real-time performance. The code is at https://github.com/ucaswangls/EfficientSCI.git.

Chingis Oinar, Miao Cao, Shanshan Fu

Mercari is the largest C2C e-commerce marketplace in Japan, having more than 20 million active monthly users. Search being the fundamental way to discover desired items, we have always had a substantial amount of data with implicit feedback. Although we actively take advantage of that to provide the best service for our users, the correlation of implicit feedback for such tasks as image quality assessment is not trivial. Many traditional lines of research in Machine Learning (ML) are similarly motivated by the insatiable appetite of Deep Learning (DL) models for well-labelled training data. Weak supervision is about leveraging higher-level and/or noisier supervision over unlabeled data. Large Language Models (LLMs) are being actively studied and used for data labelling tasks. We present how we leverage a Chain-of-Thought (CoT) to enable LLM to produce image aesthetics labels that correlate well with human behavior in e-commerce settings. Leveraging LLMs is more cost-effective compared to explicit human judgment, while significantly improving the explainability of deep image quality evaluation which is highly important for customer journey optimization at Mercari. We propose a cost-efficient LLM-driven approach for assessing and predicting image quality in e-commerce settings, which is very convenient for proof-of-concept testing. We show that our LLM-produced labels correlate with user behavior on Mercari. Finally, we show our results from an online experimentation, where we achieved a significant growth in sales on the web platform.

Xiongfei Su, Siyuan Li, Yuning Cui et al.

Image dehazing is a crucial task that involves the enhancement of degraded images to recover their sharpness and textures. While vision Transformers have exhibited impressive results in diverse dehazing tasks, their quadratic complexity and lack of dehazing priors pose significant drawbacks for real-world applications. In this paper, guided by triple priors, Bright Channel Prior (BCP), Dark Channel Prior (DCP), and Histogram Equalization (HE), we propose a \textit{P}rior-\textit{g}uided Hierarchical \textit{H}armonization Network (PGH$^2$Net) for image dehazing. PGH$^2$Net is built upon the UNet-like architecture with an efficient encoder and decoder, consisting of two module types: (1) Prior aggregation module that injects B/DCP and selects diverse contexts with gating attention. (2) Feature harmonization modules that subtract low-frequency components from spatial and channel aspects and learn more informative feature distributions to equalize the feature maps.

Andre Rusli, Miao Cao, Shoma Ishimoto et al.

Consumer-to-consumer (C2C) marketplaces pose distinct retrieval challenges: short, ambiguous queries; noisy, user-generated listings; and strict production constraints. This paper reports our experiment to build a domain-aware Japanese text-embedding approach to improve the quality of search at Mercari, Japan's largest C2C marketplace. We experimented with fine-tuning on purchase-driven query-title pairs, using role-specific prefixes to model query-item asymmetry. To meet production constraints, we apply Matryoshka Representation Learning to obtain compact, truncation-robust embeddings. Offline evaluation on historical search logs shows consistent gains over a strong generic encoder, with particularly large improvements when replacing PCA compression with Matryoshka truncation. A manual assessment further highlights better handling of proper nouns, marketplace-specific semantics, and term-importance alignment. Additionally, an initial online A/B test demonstrates statistically significant improvements in revenue per user and search-flow efficiency, with transaction frequency maintained. Results show that domain-aware embeddings improve relevance and efficiency at scale and form a practical foundation for richer LLM-era search experiences.

Miao Cao, Siming Zheng, Lishun Wang et al.

Digital cameras consume ~0.1 microjoule per pixel to capture and encode video, resulting in a power usage of ~20W for a 4K sensor operating at 30 fps. Imagining gigapixel cameras operating at 100-1000 fps, the current processing model is unsustainable. To address this, physical layer compressive measurement has been proposed to reduce power consumption per pixel by 10-100X. Video Snapshot Compressive Imaging (SCI) introduces high frequency modulation in the optical sensor layer to increase effective frame rate. A commonly used sampling strategy of video SCI is Random Sampling (RS) where each mask element value is randomly set to be 0 or 1. Similarly, image inpainting (I2P) has demonstrated that images can be recovered from a fraction of the image pixels. Inspired by I2P, we propose Ultra-Sparse Sampling (USS) regime, where at each spatial location, only one sub-frame is set to 1 and all others are set to 0. We then build a Digital Micro-mirror Device (DMD) encoding system to verify the effectiveness of our USS strategy. Ideally, we can decompose the USS measurement into sub-measurements for which we can utilize I2P algorithms to recover high-speed frames. However, due to the mismatch between the DMD and CCD, the USS measurement cannot be perfectly decomposed. To this end, we propose BSTFormer, a sparse TransFormer that utilizes local Block attention, global Sparse attention, and global Temporal attention to exploit the sparsity of the USS measurement. Extensive results on both simulated and real-world data show that our method significantly outperforms all previous state-of-the-art algorithms. Additionally, an essential advantage of the USS strategy is its higher dynamic range than that of the RS strategy. Finally, from the application perspective, the USS strategy is a good choice to implement a complete video SCI system on chip due to its fixed exposure time.

Haomiao Zhang, Zhangyuan Li, Yanling Piao et al.

Computer-generated holography (CGH) has gained wide attention with deep learning-based algorithms. However, due to its nonlinear and ill-posed nature, challenges remain in achieving accurate and stable reconstruction. Specifically, ($i$) the widely used end-to-end networks treat the reconstruction model as a black box, ignoring underlying physical relationships, which reduces interpretability and flexibility. ($ii$) CNN-based CGH algorithms have limited receptive fields, hindering their ability to capture long-range dependencies and global context. ($iii$) Angular spectrum method (ASM)-based models are constrained to finite near-fields.In this paper, we propose a Deep Unfolding Network (DUN) that decomposes gradient descent into two modules: an adaptive bandwidth-preserving model (ABPM) and a phase-domain complex-valued denoiser (PCD), providing more flexibility. ABPM allows for wider working distances compared to ASM-based methods. At the same time, PCD leverages its complex-valued deformable self-attention module to capture global features and enhance performance, achieving a PSNR over 35 dB. Experiments on simulated and real data show state-of-the-art results.

Haomiao Zhang, Miao Cao, Xuan Yu et al.

Computer-generated holography (CGH) is a promising technology for next-generation displays. However, generating high-speed, high-quality holographic video requires both high frame rate display and efficient computation, but is constrained by two key limitations: ($i$) Learning-based models often produce over-smoothed phases with narrow angular spectra, causing severe color crosstalk in high frame rate full-color displays such as depth-division multiplexing and thus resulting in a trade-off between frame rate and color fidelity. ($ii$) Existing frame-by-frame optimization methods typically optimize frames independently, neglecting spatial-temporal correlations between consecutive frames and leading to computationally inefficient solutions. To overcome these challenges, in this paper, we propose a novel high-speed full-color video CGH generation scheme. First, we introduce Spectrum-Guided Depth Division Multiplexing (SGDDM), which optimizes phase distributions via frequency modulation, enabling high-fidelity full-color display at high frame rates. Second, we present HoloMamba, a lightweight asymmetric Mamba-Unet architecture that explicitly models spatial-temporal correlations across video sequences to enhance reconstruction quality and computational efficiency. Extensive simulated and real-world experiments demonstrate that SGDDM achieves high-fidelity full-color display without compromise in frame rate, while HoloMamba generates FHD (1080p) full-color holographic video at over 260 FPS, more than 2.6$\times$ faster than the prior state-of-the-art Divide-Conquer-and-Merge Strategy.

Yutong Du, Zicheng Liu, Miao Cao et al.

Deep neural networks have been applied to address electromagnetic inverse scattering problems (ISPs) and shown superior imaging performances, which can be affected by the training dataset, the network architecture and the applied loss function. Here, the quality of data samples is cared and valued by the defined quality factor. Based on the quality factor, the composition of the training dataset is optimized. The network architecture is integrated with the residual connections and channel attention mechanism to improve feature extraction. A loss function that incorporates data-fitting error, physical-information constraints and the desired feature of the solution is designed and analyzed to suppress the background artifacts and improve the reconstruction accuracy. Various numerical analysis are performed to demonstrate the superiority of the proposed quality-factor inspired deep neural network (QuaDNN) solver and the imaging performance is finally verified by experimental imaging test.

Jincheng Yang, Lishun Wang, Miao Cao et al.

We study the inverse problem of Coded Aperture Snapshot Spectral Imaging (CASSI), which captures a spatial-spectral data cube using snapshot 2D measurements and uses algorithms to reconstruct 3D hyperspectral images (HSI). However, current methods based on Convolutional Neural Networks (CNNs) struggle to capture long-range dependencies and non-local similarities. The recently popular Transformer-based methods are poorly deployed on downstream tasks due to the high computational cost caused by self-attention. In this paper, we propose Coarse-Fine Spectral-Aware Deformable Convolution Network (CFSDCN), applying deformable convolutional networks (DCN) to this task for the first time. Considering the sparsity of HSI, we design a deformable convolution module that exploits its deformability to capture long-range dependencies and non-local similarities. In addition, we propose a new spectral information interaction module that considers both coarse-grained and fine-grained spectral similarities. Extensive experiments demonstrate that our CFSDCN significantly outperforms previous state-of-the-art (SOTA) methods on both simulated and real HSI datasets.

Cenyuan Zhang, Xiaoqing Zheng, Ruicheng Yin et al.

Deciphering natural language from brain activity through non-invasive devices remains a formidable challenge. Previous non-invasive decoders either require multiple experiments with identical stimuli to pinpoint cortical regions and enhance signal-to-noise ratios in brain activity, or they are limited to discerning basic linguistic elements such as letters and words. We propose a novel approach to decoding continuous language from single-trial non-invasive fMRI recordings, in which a three-dimensional convolutional network augmented with information bottleneck is developed to automatically identify responsive voxels to stimuli, and a character-based decoder is designed for the semantic reconstruction of continuous language characterized by inherent character structures. The resulting decoder can produce intelligible textual sequences that faithfully capture the meaning of perceived speech both within and across subjects, while existing decoders exhibit significantly inferior performance in cross-subject contexts. The ability to decode continuous language from single trials across subjects demonstrates the promising applications of non-invasive language brain-computer interfaces in both healthcare and neuroscience.

Miao Cao, Satoshi Ikehata, Kiyoharu Aizawa

360° cameras have gained popularity over the last few years. In this paper, we propose two fundamental techniques -- Field-of-View IoU (FoV-IoU) and 360Augmentation for object detection in 360° images. Although most object detection neural networks designed for the perspective images are applicable to 360° images in equirectangular projection (ERP) format, their performance deteriorates owing to the distortion in ERP images. Our method can be readily integrated with existing perspective object detectors and significantly improves the performance. The FoV-IoU computes the intersection-over-union of two Field-of-View bounding boxes in a spherical image which could be used for training, inference, and evaluation while 360Augmentation is a data augmentation technique specific to 360° object detection task which randomly rotates a spherical image and solves the bias due to the sphere-to-plane projection. We conduct extensive experiments on the 360indoor dataset with different types of perspective object detectors and show the consistent effectiveness of our method.

Dongwei Jiang, Wubo Li, Miao Cao et al.

Self-supervised visual pretraining has shown significant progress recently. Among those methods, SimCLR greatly advanced the state of the art in self-supervised and semi-supervised learning on ImageNet. The input feature representations for speech and visual tasks are both continuous, so it is natural to consider applying similar objective on speech representation learning. In this paper, we propose Speech SimCLR, a new self-supervised objective for speech representation learning. During training, Speech SimCLR applies augmentation on raw speech and its spectrogram. Its objective is the combination of contrastive loss that maximizes agreement between differently augmented samples in the latent space and reconstruction loss of input representation. The proposed method achieved competitive results on speech emotion recognition and speech recognition.

Dongwei Jiang, Wubo Li, Ruixiong Zhang et al.

Building a good speech recognition system usually requires large amounts of transcribed data, which is expensive to collect. To tackle this problem, many unsupervised pre-training methods have been proposed. Among these methods, Masked Predictive Coding achieved significant improvements on various speech recognition datasets with BERT-like Masked Reconstruction loss and Transformer backbone. However, many aspects of MPC have not been fully investigated. In this paper, we conduct a further study on MPC and focus on three important aspects: the effect of pre-training data speaking style, its extension on streaming model, and how to better transfer learned knowledge from pre-training stage to downstream tasks. Experiments reveled that pre-training data with matching speaking style is more useful on downstream recognition tasks. A unified training objective with APC and MPC provided 8.46% relative error reduction on streaming model trained on HKUST. Also, the combination of target data adaption and layer-wise discriminative training helped the knowledge transfer of MPC, which achieved 3.99% relative error reduction on AISHELL over a strong baseline.