Zheng Li

Zheng Li, Caili Guo, Zerun Feng et al.

Most image-text retrieval work adopts binary labels indicating whether a pair of image and text matches or not. Such a binary indicator covers only a limited subset of image-text semantic relations, which is insufficient to represent relevance degrees between images and texts described by continuous labels such as image captions. The visual-semantic embedding space obtained by learning binary labels is incoherent and cannot fully characterize the relevance degrees. In addition to the use of binary labels, this paper further incorporates continuous pseudo labels (generally approximated by text similarity between captions) to indicate the relevance degrees. To learn a coherent embedding space, we propose an image-text retrieval framework with Binary and Continuous Label Supervision (BCLS), where binary labels are used to guide the retrieval model to learn limited binary correlations, and continuous labels are complementary to the learning of image-text semantic relations. For the learning of binary labels, we improve the common Triplet ranking loss with Soft Negative mining (Triplet-SN) to improve convergence. For the learning of continuous labels, we design Kendall ranking loss inspired by Kendall rank correlation coefficient (Kendall), which improves the correlation between the similarity scores predicted by the retrieval model and the continuous labels. To mitigate the noise introduced by the continuous pseudo labels, we further design Sliding Window sampling and Hard Sample mining strategy (SW-HS) to alleviate the impact of noise and reduce the complexity of our framework to the same order of magnitude as the triplet ranking loss. Extensive experiments on two image-text retrieval benchmarks demonstrate that our method can improve the performance of state-of-the-art image-text retrieval models.

Zheng Li, Caili Guo, Xin Wang et al.

There are two popular loss functions used for vision-language retrieval, i.e., triplet loss and contrastive learning loss, both of them essentially minimize the difference between the similarities of negative pairs and positive pairs. More specifically, Triplet loss with Hard Negative mining (Triplet-HN), which is widely used in existing retrieval models to improve the discriminative ability, is easy to fall into local minima in training. On the other hand, Vision-Language Contrastive learning loss (VLC), which is widely used in the vision-language pre-training, has been shown to achieve significant performance gains on vision-language retrieval, but the performance of fine-tuning with VLC on small datasets is not satisfactory. This paper proposes a unified loss of pair similarity optimization for vision-language retrieval, providing a powerful tool for understanding existing loss functions. Our unified loss includes the hard sample mining strategy of VLC and introduces the margin used by the triplet loss for better similarity separation. It is shown that both Triplet-HN and VLC are special forms of our unified loss. Compared with the Triplet-HN, our unified loss has a fast convergence speed. Compared with the VLC, our unified loss is more discriminative and can provide better generalization in downstream fine-tuning tasks. Experiments on image-text and video-text retrieval benchmarks show that our unified loss can significantly improve the performance of the state-of-the-art retrieval models.

Zheng Li, Caili Guo, Xin Wang et al.

Recently, a series of Image-Text Matching (ITM) methods achieve impressive performance. However, we observe that most existing ITM models suffer from gradients vanishing at the beginning of training, which makes these models prone to falling into local minima. Most ITM models adopt triplet loss with Hard Negative mining (HN) as the optimization objective. We find that optimizing an ITM model using only the hard negative samples can easily lead to gradient vanishing. In this paper, we derive the condition under which the gradient vanishes during training. When the difference between the positive pair similarity and the negative pair similarity is close to 0, the gradients on both the image and text encoders will approach 0. To alleviate the gradient vanishing problem, we propose a Selectively Hard Negative Mining (SelHN) strategy, which chooses whether to mine hard negative samples according to the gradient vanishing condition. SelHN can be plug-and-play applied to existing ITM models to give them better training behavior. To further ensure the back-propagation of gradients, we construct a Residual Visual Semantic Embedding model with SelHN, denoted as RVSE++. Extensive experiments on two ITM benchmarks demonstrate the strength of RVSE++, achieving state-of-the-art performance.

Yixin Yang, Zheng Li, Qingxiu Dong et al. · pku

Understanding the deep semantics of images is essential in the era dominated by social media. However, current research works primarily on the superficial description of images, revealing a notable deficiency in the systematic investigation of the inherent deep semantics. In this work, we introduce DEEPEVAL, a comprehensive benchmark to assess Large Multimodal Models' (LMMs) capacities of visual deep semantics. DEEPEVAL includes human-annotated dataset and three progressive subtasks: fine-grained description selection, in-depth title matching, and deep semantics understanding. Utilizing DEEPEVAL, we evaluate 9 open-source LMMs and GPT-4V(ision). Our evaluation demonstrates a substantial gap between the deep semantic comprehension capabilities of existing LMMs and humans. For example, GPT-4V is 30% behind humans in understanding deep semantics, even though it achieves human-comparable performance in image description. Further analysis reveals that LMM performance on DEEPEVAL varies according to the specific facets of deep semantics explored, indicating the fundamental challenges remaining in developing LMMs.

Chun Liu, Suqiang Ma, Zheng Li et al.

Zero-shot classification of image scenes which can recognize the image scenes that are not seen in the training stage holds great promise of lowering the dependence on large numbers of labeled samples. To address the zero-shot image scene classification, the cross-modal feature alignment methods have been proposed in recent years. These methods mainly focus on matching the visual features of each image scene with their corresponding semantic descriptors in the latent space. Less attention has been paid to the contrastive relationships between different image scenes and different semantic descriptors. In light of the challenge of large intra-class difference and inter-class similarity among image scenes and the potential noisy samples, these methods are susceptible to the influence of the instances which are far from these of the same classes and close to these of other classes. In this work, we propose a multi-level cross-modal feature alignment method via contrastive learning for zero-shot classification of remote sensing image scenes. While promoting the single-instance level positive alignment between each image scene with their corresponding semantic descriptors, the proposed method takes the cross-instance contrastive relationships into consideration,and learns to keep the visual and semantic features of different classes in the latent space apart from each other. Extensive experiments have been done to evaluate the performance of the proposed method. The results show that our proposed method outperforms state of the art methods for zero-shot remote sensing image scene classification. All the code and data are available at github https://github.com/masuqiang/MCFA-Pytorch

Mingyang Song, Mao Zheng, Zheng Li et al.

Improving training efficiency continues to be one of the primary challenges in large-scale Reinforcement Learning (RL). In this paper, we investigate how context length and the complexity of training data influence the RL scaling training process of R1-distilled reasoning models, e.g., DeepSeek-R1-Distill-Qwen-1.5B. Our experimental results reveal that: (1) simply controlling the context length and curating the training data based on the input prompt length can effectively improve the training efficiency of RL scaling, achieving better performance with more concise CoT; (2) properly scaling the context length helps mitigate entropy collapse; and (3) carefully choosing the context length facilitates achieving efficient LLM training and reasoning. Inspired by these insights, we propose FastCuRL, a curriculum RL framework with stage-wise context scaling to achieve efficient LLM training and reasoning. Extensive experimental results demonstrate that FastCuRL-1.5B-V3 significantly outperforms state-of-the-art reasoning models on five competition-level benchmarks and achieves 49.6% accuracy on AIME 2024. Furthermore, FastCuRL-1.5B-Preview surpasses DeepScaleR-1.5B-Preview on five benchmarks while only using a single node with 8 GPUs and a total of 50% of training steps.

Zheng Li, Mao Zheng, Mingyang Song et al.

Recently, deep reasoning large language models(LLMs) like DeepSeek-R1 have made significant progress in tasks such as mathematics and coding. Inspired by this, several studies have employed reinforcement learning(RL) to enhance models' deep reasoning capabilities and improve machine translation(MT) quality. However, the terminology translation, an essential task in MT, remains unexplored in deep reasoning LLMs. In this paper, we propose \textbf{TAT-R1}, a terminology-aware translation model trained with reinforcement learning and word alignment. Specifically, we first extract the keyword translation pairs using a word alignment model. Then we carefully design three types of rule-based alignment rewards with the extracted alignment relationships. With those alignment rewards, the RL-trained translation model can learn to focus on the accurate translation of key information, including terminology in the source text. Experimental results show the effectiveness of TAT-R1. Our model significantly improves terminology translation accuracy compared to the baseline models while maintaining comparable performance on general translation tasks. In addition, we conduct detailed ablation studies of the DeepSeek-R1-like training paradigm for machine translation and reveal several key findings.

Shi Pu, Yijiang He, Zheng Li et al.

Facilitated by deep neural networks, video recommendation systems have made significant advances. Existing video recommendation systems directly exploit features from different modalities (e.g., user personal data, user behavior data, video titles, video tags, and visual contents) to input deep neural networks, while expecting the networks to online mine user-preferred topics implicitly from these features. However, the features lacking semantic topic information limits accurate recommendation generation. In addition, feature crosses using visual content features generate high dimensionality features that heavily downgrade the online computational efficiency of networks. In this paper, we explicitly separate topic generation from recommendation generation, propose a multimodal topic learning algorithm to exploit three modalities (i.e., tags, titles, and cover images) for generating video topics offline. The topics generated by the proposed algorithm serve as semantic topic features to facilitate preference scope determination and recommendation generation. Furthermore, we use the semantic topic features instead of visual content features to effectively reduce online computational cost. Our proposed algorithm has been deployed in the Kuaibao information streaming platform. Online and offline evaluation results show that our proposed algorithm performs favorably.

Zerun Feng, Zhimin Zeng, Caili Guo et al.

Video retrieval is a challenging research topic bridging the vision and language areas and has attracted broad attention in recent years. Previous works have been devoted to representing videos by directly encoding from frame-level features. In fact, videos consist of various and abundant semantic relations to which existing methods pay less attention. To address this issue, we propose a Visual Semantic Enhanced Reasoning Network (ViSERN) to exploit reasoning between frame regions. Specifically, we consider frame regions as vertices and construct a fully-connected semantic correlation graph. Then, we perform reasoning by novel random walk rule-based graph convolutional networks to generate region features involved with semantic relations. With the benefit of reasoning, semantic interactions between regions are considered, while the impact of redundancy is suppressed. Finally, the region features are aggregated to form frame-level features for further encoding to measure video-text similarity. Extensive experiments on two public benchmark datasets validate the effectiveness of our method by achieving state-of-the-art performance due to the powerful semantic reasoning.

Zheng Li, Ge Han, Yunqing Wei et al.

Steganography refers to the art of concealing secret messages within multiple media carriers so that an eavesdropper is unable to detect the presence and content of the hidden messages. In this paper, we firstly propose a novel key-dependent steganographic scheme that achieves steganographic objectives with adversarial training. Symmetric (secret-key) and Asymmetric (public-key) steganographic scheme are separately proposed and each scheme is successfully designed and implemented. We show that these encodings produced by our scheme improve the invisibility by 20% than previous deep-leanring-based work, and further that perform competitively remarkable undetectability 25% better than classic steganographic algorithms. Finally, we simulated our scheme in a real situation where the decoder achieved an accuracy of more than 98% of the original message.