FINE: Factorizing Knowledge for Initialization of Variable-sized Diffusion Models

Diffusion models often face slow convergence, and existing efficient training techniques, such as Parameter-Efficient Fine-Tuning (PEFT), are primarily designed for fine-tuning pre-trained models. However, these methods are limited in adapting models to variable sizes for real-world deployment, where no corresponding pre-trained models exist. To address this, we introduce FINE, a method based on the Learngene framework, to initializing downstream networks leveraging pre-trained models, while considering both model sizes and task-specific requirements. FINE decomposes pre-trained knowledge into the product of matrices (i.e., $U$, $\Sigma$, and $V$), where $U$ and $V$ are shared across network blocks as ``learngenes'', and $\Sigma$ remains layer-specific. During initialization, FINE trains only $\Sigma$ using a small subset of data, while keeping the learngene parameters fixed, marking it the first approach to integrate both size and task considerations in initialization. We provide a comprehensive benchmark for learngene-based methods in image generation tasks, and extensive experiments demonstrate that FINE consistently outperforms direct pre-training, particularly for smaller models, achieving state-of-the-art results across variable model sizes. FINE also offers significant computational and storage savings, reducing training steps by approximately $3N\times$ and storage by $5\times$, where $N$ is the number of models. Additionally, FINE's adaptability to tasks yields an average performance improvement of 4.29 and 3.30 in FID and sFID across multiple downstream datasets, highlighting its versatility and efficiency.

Joint Beamforming and Illumination Pattern Design for Beam-Hopping LEO Satellite Communications

Since hybrid beamforming (HBF) can approach the performance of fully-digital beamforming (FDBF) with much lower hardware complexity, we investigate the HBF design for beam-hopping (BH) low earth orbit (LEO) satellite communications (SatComs). Aiming at maximizing the sum-rate of totally illuminated beam positions during the whole BH period, we consider joint beamforming and illumination pattern design subject to the HBF constraints and sum-rate requirements. To address the non-convexity of the HBF constraints, we temporarily replace the HBF constraints with the FDBF constraints. Then we propose an FDBF and illumination pattern random search (FDBF-IPRS) scheme to optimize illumination patterns and fully-digital beamformers using constrained random search and fractional programming methods. To further reduce the computational complexity, we propose an FDBF and illumination pattern alternating optimization (FDBF-IPAO) scheme, where we relax the integer illumination pattern to continuous variables and after finishing all the iterations we quantize the continuous variables into integer ones. Based on the fully-digital beamformers designed by the FDBF-IPRS or FDBF-IPAO scheme, we propose an HBF alternating minimization algorithm to design the hybrid beamformers. Simulation results show that the proposed schemes can achieve satisfactory sum-rate performance for BH LEO SatComs.

Qihoo-T2X: An Efficiency-Focused Diffusion Transformer via Proxy Tokens for Text-to-Any-Task

The global self-attention mechanism in diffusion transformers involves redundant computation due to the sparse and redundant nature of visual information, and the attention map of tokens within a spatial window shows significant similarity. To address this redundancy, we propose the Proxy Token Diffusion Transformer (PT-DiT), which employs sparse representative token attention (where the number of representative tokens is much smaller than the total number of tokens) to model global visual information efficiently. Specifically, in each transformer block, we randomly sample one token from each spatial-temporal window to serve as a proxy token for that region. The global semantics are captured through the self-attention of these proxy tokens and then injected into all latent tokens via cross-attention. Simultaneously, we introduce window and shift window attention to address the limitations in detail modeling caused by the sparse attention mechanism. Building on the well-designed PT-DiT, we further develop the Qihoo-T2X family, which includes a variety of models for T2I, T2V, and T2MV tasks. Experimental results show that PT-DiT achieves competitive performance while reducing the computational complexity in both image and video generation tasks (e.g., a 48% reduction compared to DiT and a 35% reduction compared to Pixart-alpha). Our source code is available at https://github.com/360CVGroup/Qihoo-T2X.

MusicMamba: A Dual-Feature Modeling Approach for Generating Chinese Traditional Music with Modal Precision

In recent years, deep learning has significantly advanced the MIDI domain, solidifying music generation as a key application of artificial intelligence. However, existing research primarily focuses on Western music and encounters challenges in generating melodies for Chinese traditional music, especially in capturing modal characteristics and emotional expression. To address these issues, we propose a new architecture, the Dual-Feature Modeling Module, which integrates the long-range dependency modeling of the Mamba Block with the global structure capturing capabilities of the Transformer Block. Additionally, we introduce the Bidirectional Mamba Fusion Layer, which integrates local details and global structures through bidirectional scanning, enhancing the modeling of complex sequences. Building on this architecture, we propose the REMI-M representation, which more accurately captures and generates modal information in melodies. To support this research, we developed FolkDB, a high-quality Chinese traditional music dataset encompassing various styles and totaling over 11 hours of music. Experimental results demonstrate that the proposed architecture excels in generating melodies with Chinese traditional music characteristics, offering a new and effective solution for music generation.

Prompt-Softbox-Prompt: A free-text Embedding Control for Image Editing

Text-driven diffusion models have achieved remarkable success in image editing, but a crucial component in these models-text embeddings-has not been fully explored. The entanglement and opacity of text embeddings present significant challenges to achieving precise image editing. In this paper, we provide a comprehensive and in-depth analysis of text embeddings in Stable Diffusion XL, offering three key insights. First, while the 'aug_embedding' captures the full semantic content of the text, its contribution to the final image generation is relatively minor. Second, 'BOS' and 'Padding_embedding' do not contain any semantic information. Lastly, the 'EOS' holds the semantic information of all words and contains the most style features. Each word embedding plays a unique role without interfering with one another. Based on these insights, we propose a novel approach for controllable image editing using a free-text embedding control method called PSP (Prompt-Softbox-Prompt). PSP enables precise image editing by inserting or adding text embeddings within the cross-attention layers and using Softbox to define and control the specific area for semantic injection. This technique allows for obejct additions and replacements while preserving other areas of the image. Additionally, PSP can achieve style transfer by simply replacing text embeddings. Extensive experimental results show that PSP achieves significant results in tasks such as object replacement, object addition, and style transfer.

FancyVideo: Towards Dynamic and Consistent Video Generation via Cross-frame Textual Guidance

Synthesizing motion-rich and temporally consistent videos remains a challenge in artificial intelligence, especially when dealing with extended durations. Existing text-to-video (T2V) models commonly employ spatial cross-attention for text control, equivalently guiding different frame generations without frame-specific textual guidance. Thus, the model's capacity to comprehend the temporal logic conveyed in prompts and generate videos with coherent motion is restricted. To tackle this limitation, we introduce FancyVideo, an innovative video generator that improves the existing text-control mechanism with the well-designed Cross-frame Textual Guidance Module (CTGM). Specifically, CTGM incorporates the Temporal Information Injector (TII), Temporal Affinity Refiner (TAR), and Temporal Feature Booster (TFB) at the beginning, middle, and end of cross-attention, respectively, to achieve frame-specific textual guidance. Firstly, TII injects frame-specific information from latent features into text conditions, thereby obtaining cross-frame textual conditions. Then, TAR refines the correlation matrix between cross-frame textual conditions and latent features along the time dimension. Lastly, TFB boosts the temporal consistency of latent features. Extensive experiments comprising both quantitative and qualitative evaluations demonstrate the effectiveness of FancyVideo. Our approach achieves state-of-the-art T2V generation results on the EvalCrafter benchmark and facilitates the synthesis of dynamic and consistent videos. The video show results can be available at https://fancyvideo.github.io/, and we will make our code and model weights publicly available.

KIND: Knowledge Integration and Diversion in Diffusion Models

Pre-trained models have become the preferred backbone due to the expansion of model parameters, with techniques like Parameter-Efficient Fine-Tuning (PEFTs) typically fixing the parameters of these models. However, pre-trained models may not always be optimal, especially when there are discrepancies between training tasks and target tasks, potentially resulting in negative transfer. To address this, we introduce \textbf{KIND}, which performs \textbf{K}nowledge \textbf{IN}tegration and \textbf{D}iversion in diffusion models. KIND first integrates knowledge by decomposing parameter matrices of models using $U$, $\Sigma$, and $V$ matrices, formally inspired by singular value decomposition (SVD). Then it explicitly partitions the components of these matrices into \textbf{learngenes} and \textbf{tailors} to condense common and class-specific knowledge, respectively, through a class gate. In this way, KIND redefines traditional pre-training methods by adjusting training objectives from maximizing model performance on current tasks to condensing transferable common knowledge, leveraging the \textit{Learngene} framework. We conduct experiments on ImageNet-1K and compare KIND with PEFT and other learngene methods. Results indicate that KIND achieves state-of-the-art performance compared to other PEFT and learngene methods. Specifically, the images generated by KIND achieves more than 6.54 and 1.07 decrease in FID and sFID on DiT-L/2, utilizing only 45.4M trainable parameters and saving at least 35.4G FLOPs in computational cost.

BeyondCT: A deep learning model for predicting pulmonary function from chest CT scans

Abstract Background: Pulmonary function tests (PFTs) and computed tomography (CT) imaging are vital in diagnosing, managing, and monitoring lung diseases. A common issue in practice is the lack of access to recorded pulmonary functions despite available chest CT scans. Purpose: To develop and validate a deep learning algorithm for predicting pulmonary function directly from chest CT scans. Methods: The development cohort came from the Pittsburgh Lung Screening Study (PLuSS) (n=3619). The validation cohort came from the Specialized Centers of Clinically Oriented Research (SCCOR) in COPD (n=662). A deep learning model called BeyondCT, combining a three-dimensional (3D) convolutional neural network (CNN) and Vision Transformer (ViT) architecture, was used to predict forced vital capacity (FVC) and forced expiratory volume in one second (FEV1) from non-contrasted inspiratory chest CT scans. A 3D CNN model without ViT was used for comparison. Subject demographics (age, gender, smoking status) were also incorporated into the model. Performance was compared to actual PFTs using mean absolute error (MAE, L), percentage error, and R square. Results: The 3D-CNN model achieved MAEs of 0.395 L and 0.383 L, percentage errors of 13.84% and 18.85%, and R square of 0.665 and 0.679 for FVC and FEV1, respectively. The BeyondCT model without demographics had MAEs of 0.362 L and 0.371 L, percentage errors of 10.89% and 14.96%, and R square of 0.719 and 0.727, respectively. Including demographics improved performance (p<0.05), with MAEs of 0.356 L and 0.353 L, percentage errors of 10.79% and 14.82%, and R square of 0.77 and 0.739 for FVC and FEV1 in the test set. Conclusion: The BeyondCT model showed robust performance in predicting lung function from non-contrast inspiratory chest CT scans.

A Labeled Ophthalmic Ultrasound Dataset with Medical Report Generation Based on Cross-modal Deep Learning

Ultrasound imaging reveals eye morphology and aids in diagnosing and treating eye diseases. However, interpreting diagnostic reports requires specialized physicians. We present a labeled ophthalmic dataset for the precise analysis and the automated exploration of medical images along with their associated reports. It collects three modal data, including the ultrasound images, blood flow information and examination reports from 2,417 patients at an ophthalmology hospital in Shenyang, China, during the year 2018, in which the patient information is de-identified for privacy protection. To the best of our knowledge, it is the only ophthalmic dataset that contains the three modal information simultaneously. It incrementally consists of 4,858 images with the corresponding free-text reports, which describe 15 typical imaging findings of intraocular diseases and the corresponding anatomical locations. Each image shows three kinds of blood flow indices at three specific arteries, i.e., nine parameter values to describe the spectral characteristics of blood flow distribution. The reports were written by ophthalmologists during the clinical care. The proposed dataset is applied to generate medical report based on the cross-modal deep learning model. The experimental results demonstrate that our dataset is suitable for training supervised models concerning cross-modal medical data.

UMono: Physical Model Informed Hybrid CNN-Transformer Framework for Underwater Monocular Depth Estimation

Underwater monocular depth estimation serves as the foundation for tasks such as 3D reconstruction of underwater scenes. However, due to the influence of light and medium, the underwater environment undergoes a distinctive imaging process, which presents challenges in accurately estimating depth from a single image. The existing methods fail to consider the unique characteristics of underwater environments, leading to inadequate estimation results and limited generalization performance. Furthermore, underwater depth estimation requires extracting and fusing both local and global features, which is not fully explored in existing methods. In this paper, an end-to-end learning framework for underwater monocular depth estimation called UMono is presented, which incorporates underwater image formation model characteristics into network architecture, and effectively utilize both local and global features of underwater image. Experimental results demonstrate that the proposed method is effective for underwater monocular depth estimation and outperforms the existing methods in both quantitative and qualitative analyses.