p-MoD: Building Mixture-of-Depths MLLMs via Progressive Ratio Decay

Despite the remarkable performance of multimodal large language models (MLLMs) across diverse tasks, the substantial training and inference costs impede their advancement. The majority of computation stems from the overwhelming volume of vision tokens processed by the transformer decoder. In this paper, we propose to build efficient MLLMs by leveraging the Mixture-of-Depths (MoD) mechanism, where each transformer decoder layer selects essential vision tokens to process while skipping redundant ones. However, integrating MoD into MLLMs is non-trivial. To address the challenges of training and inference stability as well as limited training data, we adapt the MoD module with two novel designs: tanh-gated weight normalization (TanhNorm) and symmetric token reweighting (STRing). Moreover, we observe that vision tokens exhibit higher redundancy in deeper layer and thus design a progressive ratio decay (PRD) strategy, which gradually reduces the token retention ratio layer by layer, employing a shifted cosine schedule. This crucial design fully unleashes the potential of MoD, significantly boosting the efficiency and performance of our models. To validate the effectiveness of our approach, we conduct extensive experiments with two baseline models across 14 benchmarks. Our model, p-MoD, matches or even surpasses the performance of the baseline models, with only 55.6% TFLOPs and 53.8% KV cache storage during inference, and 77.7% GPU hours during training.

CamI2V: Camera-Controlled Image-to-Video Diffusion Model

Recently, camera pose, as a user-friendly and physics-related condition, has been introduced into text-to-video diffusion model for camera control. However, existing methods simply inject camera conditions through a side input. These approaches neglect the inherent physical knowledge of camera pose, resulting in imprecise camera control, inconsistencies, and also poor interpretability. In this paper, we emphasize the necessity of integrating explicit physical constraints into model design. Epipolar attention is proposed for modeling all cross-frame relationships from a novel perspective of noised condition. This ensures that features are aggregated from corresponding epipolar lines in all noised frames, overcoming the limitations of current attention mechanisms in tracking displaced features across frames, especially when features move significantly with the camera and become obscured by noise. Additionally, we introduce register tokens to handle cases without intersections between frames, commonly caused by rapid camera movements, dynamic objects, or occlusions. To support image-to-video, we propose the multiple guidance scale to allow for precise control for image, text, and camera, respectively. Furthermore, we establish a more robust and reproducible evaluation pipeline to solve the inaccuracy and instability of existing camera control measurement. We achieve a 25.5\% improvement in camera controllability on RealEstate10K while maintaining strong generalization to out-of-domain images. Only 24GB and 12GB are required for training and inference, respectively. We plan to release checkpoints, along with training and evaluation codes. Dynamic videos are best viewed at \url{https://zgctroy.github.io/CamI2V}.

CustomCrafter: Customized Video Generation with Preserving Motion and Concept Composition Abilities

Customized video generation aims to generate high-quality videos guided by text prompts and subject's reference images. However, since it is only trained on static images, the fine-tuning process of subject learning disrupts abilities of video diffusion models (VDMs) to combine concepts and generate motions. To restore these abilities, some methods use additional video similar to the prompt to fine-tune or guide the model. This requires frequent changes of guiding videos and even re-tuning of the model when generating different motions, which is very inconvenient for users. In this paper, we propose CustomCrafter, a novel framework that preserves the model's motion generation and conceptual combination abilities without additional video and fine-tuning to recovery. For preserving conceptual combination ability, we design a plug-and-play module to update few parameters in VDMs, enhancing the model's ability to capture the appearance details and the ability of concept combinations for new subjects. For motion generation, we observed that VDMs tend to restore the motion of video in the early stage of denoising, while focusing on the recovery of subject details in the later stage. Therefore, we propose Dynamic Weighted Video Sampling Strategy. Using the pluggability of our subject learning modules, we reduce the impact of this module on motion generation in the early stage of denoising, preserving the ability to generate motion of VDMs. In the later stage of denoising, we restore this module to repair the appearance details of the specified subject, thereby ensuring the fidelity of the subject's appearance. Experimental results show that our method has a significant improvement compared to previous methods.

Semantic Alignment for Multimodal Large Language Models

Research on Multi-modal Large Language Models (MLLMs) towards the multi-image cross-modal instruction has received increasing attention and made significant progress, particularly in scenarios involving closely resembling images (e.g., change captioning). Existing MLLMs typically follow a two-step process in their pipelines: first, extracting visual tokens independently for each input image, and then aligning these visual tokens from different images with the Large Language Model (LLM) in its textual feature space. However, the independent extraction of visual tokens for each image may result in different semantics being prioritized for different images in the first step, leading to a lack of preservation of linking information among images for subsequent LLM analysis. This issue becomes more serious in scenarios where significant variations exist among the images (e.g., visual storytelling). To address this challenge, we introduce Semantic Alignment for Multi-modal large language models (SAM). By involving the bidirectional semantic guidance between different images in the visual-token extraction process, SAM aims to enhance the preservation of linking information for coherent analysis and align the semantics of different images before feeding them into LLM. As the test bed, we propose a large-scale dataset named MmLINK consisting of 69K samples. Different from most existing datasets for MLLMs fine-tuning, our MmLINK dataset comprises multi-modal instructions with significantly diverse images. Extensive experiments on the group captioning task and the storytelling task prove the effectiveness of our SAM model, surpassing the state-of-the-art methods by a large margin (+37% for group captioning and +22% for storytelling on CIDEr score). Project page: https://mccartney01.github.io/SAM.

MacFormer: Semantic Segmentation with Fine Object Boundaries

Semantic segmentation involves assigning a specific category to each pixel in an image. While Vision Transformer-based models have made significant progress, current semantic segmentation methods often struggle with precise predictions in localized areas like object boundaries. To tackle this challenge, we introduce a new semantic segmentation architecture, ``MacFormer'', which features two key components. Firstly, using learnable agent tokens, a Mutual Agent Cross-Attention (MACA) mechanism effectively facilitates the bidirectional integration of features across encoder and decoder layers. This enables better preservation of low-level features, such as elementary edges, during decoding. Secondly, a Frequency Enhancement Module (FEM) in the decoder leverages high-frequency and low-frequency components to boost features in the frequency domain, benefiting object boundaries with minimal computational complexity increase. MacFormer is demonstrated to be compatible with various network architectures and outperforms existing methods in both accuracy and efficiency on benchmark datasets ADE20K and Cityscapes under different computational constraints.

HAFormer: Unleashing the Power of Hierarchy-Aware Features for Lightweight Semantic Segmentation

Both Convolutional Neural Networks (CNNs) and Transformers have shown great success in semantic segmentation tasks. Efforts have been made to integrate CNNs with Transformer models to capture both local and global context interactions. However, there is still room for enhancement, particularly when considering constraints on computational resources. In this paper, we introduce HAFormer, a model that combines the hierarchical features extraction ability of CNNs with the global dependency modeling capability of Transformers to tackle lightweight semantic segmentation challenges. Specifically, we design a Hierarchy-Aware Pixel-Excitation (HAPE) module for adaptive multi-scale local feature extraction. During the global perception modeling, we devise an Efficient Transformer (ET) module streamlining the quadratic calculations associated with traditional Transformers. Moreover, a correlation-weighted Fusion (cwF) module selectively merges diverse feature representations, significantly enhancing predictive accuracy. HAFormer achieves high performance with minimal computational overhead and compact model size, achieving 74.2% mIoU on Cityscapes and 71.1% mIoU on CamVid test datasets, with frame rates of 105FPS and 118FPS on a single 2080Ti GPU. The source codes are available at https://github.com/XU-GITHUB-curry/HAFormer.

Explainable AI Security: Exploring Robustness of Graph Neural Networks to Adversarial Attacks

Graph neural networks (GNNs) have achieved tremendous success, but recent studies have shown that GNNs are vulnerable to adversarial attacks, which significantly hinders their use in safety-critical scenarios. Therefore, the design of robust GNNs has attracted increasing attention. However, existing research has mainly been conducted via experimental trial and error, and thus far, there remains a lack of a comprehensive understanding of the vulnerability of GNNs. To address this limitation, we systematically investigate the adversarial robustness of GNNs by considering graph data patterns, model-specific factors, and the transferability of adversarial examples. Through extensive experiments, a set of principled guidelines is obtained for improving the adversarial robustness of GNNs, for example: (i) rather than highly regular graphs, the training graph data with diverse structural patterns is crucial for model robustness, which is consistent with the concept of adversarial training; (ii) the large model capacity of GNNs with sufficient training data has a positive effect on model robustness, and only a small percentage of neurons in GNNs are affected by adversarial attacks; (iii) adversarial transfer is not symmetric and the adversarial examples produced by the small-capacity model have stronger adversarial transferability. This work illuminates the vulnerabilities of GNNs and opens many promising avenues for designing robust GNNs.

GraphMU: Repairing Robustness of Graph Neural Networks via Machine Unlearning

Graph Neural Networks (GNNs) have demonstrated significant application potential in various fields. However, GNNs are still vulnerable to adversarial attacks. Numerous adversarial defense methods on GNNs are proposed to address the problem of adversarial attacks. However, these methods can only serve as a defense before poisoning, but cannot repair poisoned GNN. Therefore, there is an urgent need for a method to repair poisoned GNN. In this paper, we address this gap by introducing the novel concept of model repair for GNNs. We propose a repair framework, Repairing Robustness of Graph Neural Networks via Machine Unlearning (GraphMU), which aims to fine-tune poisoned GNN to forget adversarial samples without the need for complete retraining. We also introduce a unlearning validation method to ensure that our approach effectively forget specified poisoned data. To evaluate the effectiveness of GraphMU, we explore three fine-tuned subgraph construction scenarios based on the available perturbation information: (i) Known Perturbation Ratios, (ii) Known Complete Knowledge of Perturbations, and (iii) Unknown any Knowledge of Perturbations. Our extensive experiments, conducted across four citation datasets and four adversarial attack scenarios, demonstrate that GraphMU can effectively restore the performance of poisoned GNN.

Open-Vocabulary Spatio-Temporal Action Detection

Spatio-temporal action detection (STAD) is an important fine-grained video understanding task. Current methods require box and label supervision for all action classes in advance. However, in real-world applications, it is very likely to come across new action classes not seen in training because the action category space is large and hard to enumerate. Also, the cost of data annotation and model training for new classes is extremely high for traditional methods, as we need to perform detailed box annotations and re-train the whole network from scratch. In this paper, we propose a new challenging setting by performing open-vocabulary STAD to better mimic the situation of action detection in an open world. Open-vocabulary spatio-temporal action detection (OV-STAD) requires training a model on a limited set of base classes with box and label supervision, which is expected to yield good generalization performance on novel action classes. For OV-STAD, we build two benchmarks based on the existing STAD datasets and propose a simple but effective method based on pretrained video-language models (VLM). To better adapt the holistic VLM for the fine-grained action detection task, we carefully fine-tune it on the localized video region-text pairs. This customized fine-tuning endows the VLM with better motion understanding, thus contributing to a more accurate alignment between video regions and texts. Local region feature and global video feature fusion before alignment is adopted to further improve the action detection performance by providing global context. Our method achieves a promising performance on novel classes.

SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos

Video-based visual relation detection tasks, such as video scene graph generation, play important roles in fine-grained video understanding. However, current video visual relation detection datasets have two main limitations that hinder the progress of research in this area. First, they do not explore complex human-human interactions in multi-person scenarios. Second, the relation types of existing datasets have relatively low-level semantics and can be often recognized by appearance or simple prior information, without the need for detailed spatio-temporal context reasoning. Nevertheless, comprehending high-level interactions between humans is crucial for understanding complex multi-person videos, such as sports and surveillance videos. To address this issue, we propose a new video visual relation detection task: video human-human interaction detection, and build a dataset named SportsHHI for it. SportsHHI contains 34 high-level interaction classes from basketball and volleyball sports. 118,075 human bounding boxes and 50,649 interaction instances are annotated on 11,398 keyframes. To benchmark this, we propose a two-stage baseline method and conduct extensive experiments to reveal the key factors for a successful human-human interaction detector. We hope that SportsHHI can stimulate research on human interaction understanding in videos and promote the development of spatio-temporal context modeling techniques in video visual relation detection.