Teaching Video Diffusion Model with Latent Physical Phenomenon Knowledge

Video diffusion models have exhibited tremendous progress in various video generation tasks. However, existing models struggle to capture latent physical knowledge, failing to infer physical phenomena that are challenging to articulate with natural language. Generating videos following the fundamental physical laws is still an opening challenge. To address this challenge, we propose a novel method to teach video diffusion models with latent physical phenomenon knowledge, enabling the accurate generation of physically informed phenomena. Specifically, we first pretrain Masked Autoencoders (MAE) to reconstruct the physical phenomena, resulting in output embeddings that encapsulate latent physical phenomenon knowledge. Leveraging these embeddings, we could generate the pseudo-language prompt features based on the aligned spatial relationships between CLIP vision and language encoders. Particularly, given that diffusion models typically use CLIP's language encoder for text prompt embeddings, our approach integrates the CLIP visual features informed by latent physical knowledge into a quaternion hidden space. This enables the modeling of spatial relationships to produce physical knowledge-informed pseudo-language prompts. By incorporating these prompt features and fine-tuning the video diffusion model in a parameter-efficient manner, the physical knowledge-informed videos are successfully generated. We validate our method extensively through both numerical simulations and real-world observations of physical phenomena, demonstrating its remarkable performance across diverse scenarios.

Deploying Multi-task Online Server with Large Language Model

In the industry, numerous tasks are deployed online. Traditional approaches often tackle each task separately by its own network, which leads to excessive costs for developing and scaling models, especially in the context of large language models. Although multi-task methods can save costs through parameter sharing, they often struggle to outperform single-task methods in real-world applications. To tackle these challenges, we present a three-stage multi-task learning framework for large language models. It involves task filtering, followed by fine-tuning on high-resource tasks, and finally fine-tuning on all tasks. We conducted comprehensive experiments in single-task and multi-task settings. Our approach, exemplified on different benchmarks, demonstrates that it is able to achieve performance comparable to the single-task method while reducing up to 90.9\% of its overhead.

A Simple Approach to Unifying Diffusion-based Conditional Generation

Recent progress in image generation has sparked research into controlling these models through condition signals, with various methods addressing specific challenges in conditional generation. Instead of proposing another specialized technique, we introduce a simple, unified framework to handle diverse conditional generation tasks involving a specific image-condition correlation. By learning a joint distribution over a correlated image pair (e.g. image and depth) with a diffusion model, our approach enables versatile capabilities via different inference-time sampling schemes, including controllable image generation (e.g. depth to image), estimation (e.g. image to depth), signal guidance, joint generation (image & depth), and coarse control. Previous attempts at unification often introduce significant complexity through multi-stage training, architectural modification, or increased parameter counts. In contrast, our simple formulation requires a single, computationally efficient training stage, maintains the standard model input, and adds minimal learned parameters (15% of the base model). Moreover, our model supports additional capabilities like non-spatially aligned and coarse conditioning. Extensive results show that our single model can produce comparable results with specialized methods and better results than prior unified methods. We also demonstrate that multiple models can be effectively combined for multi-signal conditional generation.

Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics

While humans effortlessly discern intrinsic dynamics and adapt to new scenarios, modern AI systems often struggle. Current methods for visual grounding of dynamics either use pure neural-network-based simulators (black box), which may violate physical laws, or traditional physical simulators (white box), which rely on expert-defined equations that may not fully capture actual dynamics. We propose the Neural Material Adaptor (NeuMA), which integrates existing physical laws with learned corrections, facilitating accurate learning of actual dynamics while maintaining the generalizability and interpretability of physical priors. Additionally, we propose Particle-GS, a particle-driven 3D Gaussian Splatting variant that bridges simulation and observed images, allowing back-propagate image gradients to optimize the simulator. Comprehensive experiments on various dynamics in terms of grounded particle accuracy, dynamic rendering quality, and generalization ability demonstrate that NeuMA can accurately capture intrinsic dynamics.

Rethinking the Principle of Gradient Smooth Methods in Model Explanation

Gradient Smoothing is an efficient approach to reducing noise in gradient-based model explanation method. SmoothGrad adds Gaussian noise to mitigate much of these noise. However, the crucial hyper-parameter in this method, the variance $\sigma$ of Gaussian noise, is set manually or with heuristic approach. However, it results in the smoothed gradients still containing a certain amount of noise. In this paper, we aim to interpret SmoothGrad as a corollary of convolution, thereby re-understanding the gradient noise and the role of $\sigma$ from the perspective of confidence level. Furthermore, we propose an adaptive gradient smoothing method, AdaptGrad, based on these insights. Through comprehensive experiments, both qualitative and quantitative results demonstrate that AdaptGrad could effectively reduce almost all the noise in vanilla gradients compared with baselines methods. AdaptGrad is simple and universal, making it applicable for enhancing gradient-based interpretability methods for better visualization.

QuadMamba: Learning Quadtree-based Selective Scan for Visual State Space Model

Recent advancements in State Space Models, notably Mamba, have demonstrated superior performance over the dominant Transformer models, particularly in reducing the computational complexity from quadratic to linear. Yet, difficulties in adapting Mamba from language to vision tasks arise due to the distinct characteristics of visual data, such as the spatial locality and adjacency within images and large variations in information granularity across visual tokens. Existing vision Mamba approaches either flatten tokens into sequences in a raster scan fashion, which breaks the local adjacency of images, or manually partition tokens into windows, which limits their long-range modeling and generalization capabilities. To address these limitations, we present a new vision Mamba model, coined QuadMamba, that effectively captures local dependencies of varying granularities via quadtree-based image partition and scan. Concretely, our lightweight quadtree-based scan module learns to preserve the 2D locality of spatial regions within learned window quadrants. The module estimates the locality score of each token from their features, before adaptively partitioning tokens into window quadrants. An omnidirectional window shifting scheme is also introduced to capture more intact and informative features across different local regions. To make the discretized quadtree partition end-to-end trainable, we further devise a sequence masking strategy based on Gumbel-Softmax and its straight-through gradient estimator. Extensive experiments demonstrate that QuadMamba achieves state-of-the-art performance in various vision tasks, including image classification, object detection, instance segmentation, and semantic segmentation. The code is in https://github.com/VISION-SJTU/QuadMamba.

PredFormer: Transformers Are Effective Spatial-Temporal Predictive Learners

Spatiotemporal predictive learning methods generally fall into two categories: recurrent-based approaches, which face challenges in parallelization and performance, and recurrent-free methods, which employ convolutional neural networks (CNNs) as encoder-decoder architectures. These methods benefit from strong inductive biases but often at the expense of scalability and generalization. This paper proposes PredFormer, a pure transformer-based framework for spatiotemporal predictive learning. Motivated by the Vision Transformers (ViT) design, PredFormer leverages carefully designed Gated Transformer blocks, following a comprehensive analysis of 3D attention mechanisms, including full-, factorized-, and interleaved- spatial-temporal attention. With its recurrent-free, transformer-based design, PredFormer is both simple and efficient, significantly outperforming previous methods by large margins. Extensive experiments on synthetic and real-world datasets demonstrate that PredFormer achieves state-of-the-art performance. On Moving MNIST, PredFormer achieves a 51.3% reduction in MSE relative to SimVP. For TaxiBJ, the model decreases MSE by 33.1% and boosts FPS from 533 to 2364. Additionally, on WeatherBench, it reduces MSE by 11.1% while enhancing FPS from 196 to 404. These performance gains in both accuracy and efficiency demonstrate PredFormer's potential for real-world applications. The source code will be released at https://github.com/yyyujintang/PredFormer.

Open-Vocabulary Remote Sensing Image Semantic Segmentation

Open-vocabulary image semantic segmentation (OVS) seeks to segment images into semantic regions across an open set of categories. Existing OVS methods commonly depend on foundational vision-language models and utilize similarity computation to tackle OVS tasks. However, these approaches are predominantly tailored to natural images and struggle with the unique characteristics of remote sensing images, such as rapidly changing orientations and significant scale variations. These challenges complicate OVS tasks in earth vision, requiring specialized approaches. To tackle this dilemma, we propose the first OVS framework specifically designed for remote sensing imagery, drawing inspiration from the distinct remote sensing traits. Particularly, to address the varying orientations, we introduce a rotation-aggregative similarity computation module that generates orientation-adaptive similarity maps as initial semantic maps. These maps are subsequently refined at both spatial and categorical levels to produce more accurate semantic maps. Additionally, to manage significant scale changes, we integrate multi-scale image features into the upsampling process, resulting in the final scale-aware semantic masks. To advance OVS in earth vision and encourage reproducible research, we establish the first open-sourced OVS benchmark for remote sensing imagery, including four public remote sensing datasets. Extensive experiments on this benchmark demonstrate our proposed method achieves state-of-the-art performance. All codes and datasets are available at https://github.com/caoql98/OVRS.

CP-VoteNet: Contrastive Prototypical VoteNet for Few-Shot Point Cloud Object Detection

Few-shot point cloud 3D object detection (FS3D) aims to identify and localise objects of novel classes from point clouds, using knowledge learnt from annotated base classes and novel classes with very few annotations. Thus far, this challenging task has been approached using prototype learning, but the performance remains far from satisfactory. We find that in existing methods, the prototypes are only loosely constrained and lack of fine-grained awareness of the semantic and geometrical correlation embedded within the point cloud space. To mitigate these issues, we propose to leverage the inherent contrastive relationship within the semantic and geometrical subspaces to learn more refined and generalisable prototypical representations. To this end, we first introduce contrastive semantics mining, which enables the network to extract discriminative categorical features by constructing positive and negative pairs within training batches. Meanwhile, since point features representing local patterns can be clustered into geometric components, we further propose to impose contrastive relationship at the primitive level. Through refined primitive geometric structures, the transferability of feature encoding from base to novel classes is significantly enhanced. The above designs and insights lead to our novel Contrastive Prototypical VoteNet (CP-VoteNet). Extensive experiments on two FS3D benchmarks FS-ScanNet and FS-SUNRGBD demonstrate that CP-VoteNet surpasses current state-of-the-art methods by considerable margins across different FS3D settings. Further ablation studies conducted corroborate the rationale and effectiveness of our designs.

Free-breathing 3D cardiac extracellular volume (ECV) mapping using a linear tangent space alignment (LTSA) model

$\textbf{Purpose:}$ To develop a new method for free-breathing 3D extracellular volume (ECV) mapping of the whole heart at 3T. $\textbf{Methods:}$ A free-breathing 3D cardiac ECV mapping method was developed at 3T. T1 mapping was performed before and after contrast agent injection using a free-breathing ECG-gated inversion-recovery sequence with spoiled gradient echo readout. A linear tangent space alignment (LTSA) model-based method was used to reconstruct high-frame-rate dynamic images from (k,t)-space data sparsely sampled along a random stack-of-stars trajectory. Joint T1 and transmit B1 estimation was performed voxel-by-voxel for pre- and post-contrast T1 mapping. To account for the time-varying T1 after contrast agent injection, a linearly time-varying T1 model was introduced for post-contrast T1 mapping. ECV maps were generated by aligning pre- and post-contrast T1 maps through affine transformation. $\textbf{Results:}$ The feasibility of the proposed method was demonstrated using in vivo studies with six healthy volunteers at 3T. We obtained 3D ECV maps at a spatial resolution of 1.9$\times$1.9$\times$4.5 $mm^{3}$ and a FOV of 308$\times$308$\times$144 $mm^{3}$, with a scan time of 10.1$\pm$1.4 and 10.6$\pm$1.6 min before and after contrast agent injection, respectively. The ECV maps and the pre- and post-contrast T1 maps obtained by the proposed method were in good agreement with the 2D MOLLI method both qualitatively and quantitatively. $\textbf{Conclusion:}$ The proposed method allows for free-breathing 3D ECV mapping of the whole heart within a practically feasible imaging time. The estimated ECV values from the proposed method were comparable to those from the existing method. $\textbf{Keywords:}$ cardiac extracellular volume (ECV) mapping, cardiac T1 mapping, linear tangent space alignment (LTSA), manifold learning