Zixun Sun

Jiancong Xie, Wenjin Wang, Zhuomeng Zhang et al.

Recent advances in Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities. However, evaluating their capacity for human-like understanding in One-Image Guides remains insufficiently explored. One-Image Guides are a visual format combining text, imagery, and symbols to present reorganized and structured information for easier comprehension, which are specifically designed for human viewing and inherently embody the characteristics of human perception and understanding. Here, we present OIG-Bench, a comprehensive benchmark focused on One-Image Guide understanding across diverse domains. To reduce the cost of manual annotation, we developed a semi-automated annotation pipeline in which multiple intelligent agents collaborate to generate preliminary image descriptions, assisting humans in constructing image-text pairs. With OIG-Bench, we have conducted a comprehensive evaluation of 29 state-of-the-art MLLMs, including both proprietary and open-source models. The results show that Qwen2.5-VL-72B performs the best among the evaluated models, with an overall accuracy of 77%. Nevertheless, all models exhibit notable weaknesses in semantic understanding and logical reasoning, indicating that current MLLMs still struggle to accurately interpret complex visual-text relationships. In addition, we also demonstrate that the proposed multi-agent annotation system outperforms all MLLMs in image captioning, highlighting its potential as both a high-quality image description generator and a valuable tool for future dataset construction. Datasets are available at https://github.com/XiejcSYSU/OIG-Bench.

Lincan Cai, Shuang Li, Wenxuan Ma et al.

Large-scale pretrained models have proven immensely valuable in handling data-intensive modalities like text and image. However, fine-tuning these models for certain specialized modalities, such as protein sequence and cosmic ray, poses challenges due to the significant modality discrepancy and scarcity of labeled data. In this paper, we propose an end-to-end method, PaRe, to enhance cross-modal fine-tuning, aiming to transfer a large-scale pretrained model to various target modalities. PaRe employs a gating mechanism to select key patches from both source and target data. Through a modality-agnostic Patch Replacement scheme, these patches are preserved and combined to construct data-rich intermediate modalities ranging from easy to hard. By gradually intermediate modality generation, we can not only effectively bridge the modality gap to enhance stability and transferability of cross-modal fine-tuning, but also address the challenge of limited data in the target modality by leveraging enriched intermediate modality data. Compared with hand-designed, general-purpose, task-specific, and state-of-the-art cross-modal fine-tuning approaches, PaRe demonstrates superior performance across three challenging benchmarks, encompassing more than ten modalities.

Tingtian Li, Zixun Sun, Xinyu Xiao

Identifying highlight moments of raw video materials is crucial for improving the efficiency of editing videos that are pervasive on internet platforms. However, the extensive work of manually labeling footage has created obstacles to applying supervised methods to videos of unseen categories. The absence of an audio modality that contains valuable cues for highlight detection in many videos also makes it difficult to use multimodal strategies. In this paper, we propose a novel model with cross-modal perception for unsupervised highlight detection. The proposed model learns representations with visual-audio level semantics from image-audio pair data via a self-reconstruction task. To achieve unsupervised highlight detection, we investigate the latent representations of the network and propose the representation activation sequence learning (RASL) module with k-point contrastive learning to learn significant representation activations. To connect the visual modality with the audio modality, we use the symmetric contrastive learning (SCL) module to learn the paired visual and audio representations. Furthermore, an auxiliary task of masked feature vector sequence (FVS) reconstruction is simultaneously conducted during pretraining for representation enhancement. During inference, the cross-modal pretrained model can generate representations with paired visual-audio semantics given only the visual modality. The RASL module is used to output the highlight scores. The experimental results show that the proposed framework achieves superior performance compared to other state-of-the-art approaches.

Tingtian Li, Zixun Sun, Haoruo Zhang et al.

Recently, the witness of the rapidly growing popularity of short videos on different Internet platforms has intensified the need for a background music (BGM) retrieval system. However, existing video-music retrieval methods only based on the visual modality cannot show promising performance regarding videos with fine-grained virtual contents. In this paper, we also investigate the widely added voice-overs in short videos and propose a novel framework to retrieve BGM for fine-grained short videos. In our framework, we use the self-attention (SA) and the cross-modal attention (CMA) modules to explore the intra- and the inter-relationships of different modalities respectively. For balancing the modalities, we dynamically assign different weights to the modal features via a fusion gate. For paring the query and the BGM embeddings, we introduce a triplet pseudo-label loss to constrain the semantics of the modal embeddings. As there are no existing virtual-content video-BGM retrieval datasets, we build and release two virtual-content video datasets HoK400 and CFM400. Experimental results show that our method achieves superior performance and outperforms other state-of-the-art methods with large margins.

Tingtian Li, Zixun Sun, Xiao Chen

Human action recognition as an important application of computer vision has been studied for decades. Among various approaches, skeleton-based methods recently attract increasing attention due to their robust and superior performance. However, existing skeleton-based methods ignore the potential action relationships between different persons, while the action of a person is highly likely to be impacted by another person especially in complex events. In this paper, we propose a novel group-skeleton-based human action recognition method in complex events. This method first utilizes multi-scale spatial-temporal graph convolutional networks (MS-G3Ds) to extract skeleton features from multiple persons. In addition to the traditional key point coordinates, we also input the key point speed values to the networks for better performance. Then we use multilayer perceptrons (MLPs) to embed the distance values between the reference person and other persons into the extracted features. Lastly, all the features are fed into another MS-G3D for feature fusion and classification. For avoiding class imbalance problems, the networks are trained with a focal loss. The proposed algorithm is also our solution for the Large-scale Human-centric Video Analysis in Complex Events Challenge. Results on the HiEve dataset show that our method can give superior performance compared to other state-of-the-art methods.

Li Wang, Zixun Sun, Wentao Yao et al.

With the increasing popularity of E-sport live, Highlight Flashback has been a critical functionality of live platforms, which aggregates the overall exciting fighting scenes in a few seconds. In this paper, we introduce a novel training strategy without any additional annotation to automatically generate highlights for game video live. Considering that the existing manual edited clips contain more highlights than long game live videos, we perform pair-wise ranking constraints across clips from edited and long live videos. A multi-stream framework is also proposed to fuse spatial, temporal as well as audio features extracted from videos. To evaluate our method, we test on long game live videos with an average length of about 15 minutes. Extensive experimental results on videos demonstrate its satisfying performance on highlights generation and effectiveness by the fusion of three streams.

Wentao Yao, Zixun Sun, Xiao Chen

In order to understand content and automatically extract labels for videos of the game "Honor of Kings", it is necessary to detect and recognize characters (called "hero") together with their camps in the game video. In this paper, we propose an efficient two-stage algorithm to detect and recognize heros in game videos. First, we detect all heros in a video frame based on blood bar template-matching method, and classify them according to their camps (self/ friend/ enemy). Then we recognize the name of each hero using one or more deep convolution neural networks. Our method needs almost no work for labelling training and testing samples in the recognition stage. Experiments show its efficiency and accuracy in the task of hero detection and recognition in game videos.

Zixun Sun, Shuang Zhao, Chengwei Zhu et al.

Online personalized news product needs a suitable cover for the article. The news cover demands to be with high image quality, and draw readers' attention at same time, which is extraordinary challenging due to the subjectivity of the task. In this paper, we assess the news cover from image clarity and object salience perspective. We propose an end-to-end multi-task learning network for image clarity assessment and semantic segmentation simultaneously, the results of which can be guided for news cover assessment. The proposed network is based on a modified DeepLabv3+ model. The network backbone is used for multiple scale spatial features exaction, followed by two branches for image clarity assessment and semantic segmentation, respectively. The experiment results show that the proposed model is able to capture important content in images and performs better than single-task learning baselines on our proposed game content based CIA dataset.