Bo Zhang

Wei Zhang, Pengye Zhang, Bo Zhang et al.

In the recommendation systems, there are multiple business domains to meet the diverse interests and needs of users, and the click-through rate(CTR) of each domain can be quite different, which leads to the demand for CTR prediction modeling for different business domains. The industry solution is to use domain-specific models or transfer learning techniques for each domain. The disadvantage of the former is that the data from other domains is not utilized by a single domain model, while the latter leverage all the data from different domains, but the fine-tuned model of transfer learning may trap the model in a local optimum of the source domain, making it difficult to fit the target domain. Meanwhile, significant differences in data quantity and feature schemas between different domains, known as domain shift, may lead to negative transfer in the process of transferring. To overcome these challenges, we propose the Collaborative Cross-Domain Transfer Learning Framework (CCTL). CCTL evaluates the information gain of the source domain on the target domain using a symmetric companion network and adjusts the information transfer weight of each source domain sample using the information flow network. This approach enables full utilization of other domain data while avoiding negative migration. Additionally, a representation enhancement network is used as an auxiliary task to preserve domain-specific features. Comprehensive experiments on both public and real-world industrial datasets, CCTL achieved SOTA score on offline metrics. At the same time, the CCTL algorithm has been deployed in Meituan, bringing 4.37% CTR and 5.43% GMV lift, which is significant to the business.

Wei Zhang, Ping Zhang, Jian Dong et al.

The effectiveness of ad creatives is greatly influenced by their visual appearance. Advertising platforms can generate ad creatives with different appearances by combining creative elements provided by advertisers. However, with the increasing number of ad creative elements, it becomes challenging to select a suitable combination from the countless possibilities. The industry's mainstream approach is to select individual creative elements independently, which often overlooks the importance of interaction between creative elements during the modeling process. In response, this paper proposes a Cross-Element Combinatorial Selection framework for multiple creative elements, termed CECS. In the encoder process, a cross-element interaction is adopted to dynamically adjust the expression of a single creative element based on the current candidate creatives. In the decoder process, the creative combination problem is transformed into a cascade selection problem of multiple creative elements. A pointer mechanism with a cascade design is used to model the associations among candidates. Comprehensive experiments on real-world datasets show that CECS achieved the SOTA score on offline metrics. Moreover, the CECS algorithm has been deployed in our industrial application, resulting in a significant 6.02% CTR and 10.37% GMV lift, which is beneficial to the business.

Junhong Gou, Bo Zhang, Li Niu et al.

The virtual try-on task refers to fitting the clothes from one image onto another portrait image. In this paper, we focus on virtual accessory try-on, which fits accessory (e.g., glasses, ties) onto a face or portrait image. Unlike clothing try-on, which relies on human silhouette as guidance, accessory try-on warps the accessory into an appropriate location and shape to generate a plausible composite image. In contrast to previous try-on methods that treat foreground (i.e., accessories) and background (i.e., human faces or bodies) equally, we propose a background-oriented network to utilize the prior knowledge of human bodies and accessories. Specifically, our approach learns the human body priors and hallucinates the target locations of specified foreground keypoints in the background. Then our approach will inject foreground information with accessory priors into the background UNet. Based on the hallucinated target locations, the warping parameters are calculated to warp the foreground. Moreover, this background-oriented network can also easily incorporate auxiliary human face/body semantic segmentation supervision to further boost performance. Experiments conducted on STRAT dataset validate the effectiveness of our proposed method.

Lingxiao Lu, Jiangtong Li, Bo Zhang et al.

The goal of image composition is merging a foreground object into a background image to obtain a realistic composite image. Recently, generative composition methods are built on large pretrained diffusion models, due to their unprecedented image generation ability. However, they are weak in preserving the foreground object details. Inspired by recent text-to-image generation customized for certain object, we propose DreamCom by treating image composition as text-guided image inpainting customized for certain object. Specifically , we finetune pretrained text-guided image inpainting model based on a few reference images containing the same object, during which the text prompt contains a special token associated with this object. Then, given a new background, we can insert this object into the background with the text prompt containing the special token. In practice, the inserted object may be adversely affected by the background, so we propose masked attention mechanisms to avoid negative background interference. Experimental results on DreamEditBench and our contributed MureCom dataset show the outstanding performance of our DreamCom.

Sifan Zhou, Liang Li, Xinyu Zhang et al.

Due to highly constrained computing power and memory, deploying 3D lidar-based detectors on edge devices equipped in autonomous vehicles and robots poses a crucial challenge. Being a convenient and straightforward model compression approach, Post-Training Quantization (PTQ) has been widely adopted in 2D vision tasks. However, applying it directly to 3D lidar-based tasks inevitably leads to performance degradation. As a remedy, we propose an effective PTQ method called LiDAR-PTQ, which is particularly curated for 3D lidar detection (both SPConv-based and SPConv-free). Our LiDAR-PTQ features three main components, \textbf{(1)} a sparsity-based calibration method to determine the initialization of quantization parameters, \textbf{(2)} a Task-guided Global Positive Loss (TGPL) to reduce the disparity between the final predictions before and after quantization, \textbf{(3)} an adaptive rounding-to-nearest operation to minimize the layerwise reconstruction error. Extensive experiments demonstrate that our LiDAR-PTQ can achieve state-of-the-art quantization performance when applied to CenterPoint (both Pillar-based and Voxel-based). To our knowledge, for the very first time in lidar-based 3D detection tasks, the PTQ INT8 model's accuracy is almost the same as the FP32 model while enjoying $3\times$ inference speedup. Moreover, our LiDAR-PTQ is cost-effective being $30\times$ faster than the quantization-aware training method. Code will be released at \url{https://github.com/StiphyJay/LiDAR-PTQ}.

Fei Wei, Xinyu Zhang, Ailing Zhang et al.

With the fast-paced development of multimodal large language models (MLLMs), we can now converse with AI systems in natural languages to understand images. However, the reasoning power and world knowledge embedded in the large language models have been much less investigated and exploited for image perception tasks. In this paper, we propose Lenna, a language-enhanced reasoning detection assistant, which utilizes the robust multimodal feature representation of MLLMs, while preserving location information for detection. This is achieved by incorporating an additional <DET> token in the MLLM vocabulary that is free of explicit semantic context but serves as a prompt for the detector to identify the corresponding position. To evaluate the reasoning capability of Lenna, we construct a ReasonDet dataset to measure its performance on reasoning-based detection. Remarkably, Lenna demonstrates outstanding performance on ReasonDet and comes with significantly low training costs. It also incurs minimal transferring overhead when extended to other tasks. Our code and model will be available at https://git.io/Lenna.

Anwen Hu, Haiyang Xu, Jiabo Ye et al.

Structure information is critical for understanding the semantics of text-rich images, such as documents, tables, and charts. Existing Multimodal Large Language Models (MLLMs) for Visual Document Understanding are equipped with text recognition ability but lack general structure understanding abilities for text-rich document images. In this work, we emphasize the importance of structure information in Visual Document Understanding and propose the Unified Structure Learning to boost the performance of MLLMs. Our Unified Structure Learning comprises structure-aware parsing tasks and multi-grained text localization tasks across 5 domains: document, webpage, table, chart, and natural image. To better encode structure information, we design a simple and effective vision-to-text module H-Reducer, which can not only maintain the layout information but also reduce the length of visual features by merging horizontal adjacent patches through convolution, enabling the LLM to understand high-resolution images more efficiently. Furthermore, by constructing structure-aware text sequences and multi-grained pairs of texts and bounding boxes for publicly available text-rich images, we build a comprehensive training set DocStruct4M to support structure learning. Finally, we construct a small but high-quality reasoning tuning dataset DocReason25K to trigger the detailed explanation ability in the document domain. Our model DocOwl 1.5 achieves state-of-the-art performance on 10 visual document understanding benchmarks, improving the SOTA performance of MLLMs with a 7B LLM by more than 10 points in 5/10 benchmarks. Our codes, models, and datasets are publicly available at https://github.com/X-PLUG/mPLUG-DocOwl/tree/main/DocOwl1.5.

Dongzhi Jiang, Renrui Zhang, Ziyu Guo et al.

Answering questions with Chain-of-Thought (CoT) has significantly enhanced the reasoning capabilities of Large Language Models (LLMs), yet its impact on Large Multimodal Models (LMMs) still lacks a systematic assessment and in-depth investigation. In this paper, we introduce MME-CoT, a specialized benchmark evaluating the CoT reasoning performance of LMMs, spanning six domains: math, science, OCR, logic, space-time, and general scenes. As the first comprehensive study in this area, we propose a thorough evaluation suite incorporating three novel metrics that assess the reasoning quality, robustness, and efficiency at a fine-grained level. Leveraging curated high-quality data and a unique evaluation strategy, we conduct an in-depth analysis of state-of-the-art LMMs, uncovering several key insights: 1) Models with reflection mechanism demonstrate a superior CoT quality, with Kimi k1.5 outperforming GPT-4o and demonstrating the highest quality results; 2) CoT prompting often degrades LMM performance on perception-heavy tasks, suggesting a potentially harmful overthinking behavior; and 3) Although the CoT quality is high, LMMs with reflection exhibit significant inefficiency in both normal response and self-correction phases. We hope MME-CoT serves as a foundation for advancing multimodal reasoning in LMMs. Project Page: https://mmecot.github.io/

Zhanpeng Zhou, Zijun Chen, Yilan Chen et al.

The pretraining-finetuning paradigm has become the prevailing trend in modern deep learning. In this work, we discover an intriguing linear phenomenon in models that are initialized from a common pretrained checkpoint and finetuned on different tasks, termed as Cross-Task Linearity (CTL). Specifically, we show that if we linearly interpolate the weights of two finetuned models, the features in the weight-interpolated model are often approximately equal to the linear interpolation of features in two finetuned models at each layer. We provide comprehensive empirical evidence supporting that CTL consistently occurs for finetuned models that start from the same pretrained checkpoint. We conjecture that in the pretraining-finetuning paradigm, neural networks approximately function as linear maps, mapping from the parameter space to the feature space. Based on this viewpoint, our study unveils novel insights into explaining model merging/editing, particularly by translating operations from the parameter space to the feature space. Furthermore, we delve deeper into the root cause for the emergence of CTL, highlighting the role of pretraining.

Jingdong Zhang, Jiayuan Fan, Peng Ye et al.

Multi-task dense prediction aims at handling multiple pixel-wise prediction tasks within a unified network simultaneously for visual scene understanding. However, cross-task feature interactions of current methods are still suffering from incomplete levels of representations, less discriminative semantics in feature participants, and inefficient pair-wise task interaction processes. To tackle these under-explored issues, we propose a novel BridgeNet framework, which extracts comprehensive and discriminative intermediate Bridge Features, and conducts interactions based on them. Specifically, a Task Pattern Propagation (TPP) module is firstly applied to ensure highly semantic task-specific feature participants are prepared for subsequent interactions, and a Bridge Feature Extractor (BFE) is specially designed to selectively integrate both high-level and low-level representations to generate the comprehensive bridge features. Then, instead of conducting heavy pair-wise cross-task interactions, a Task-Feature Refiner (TFR) is developed to efficiently take guidance from bridge features and form final task predictions. To the best of our knowledge, this is the first work considering the completeness and quality of feature participants in cross-task interactions. Extensive experiments are conducted on NYUD-v2, Cityscapes and PASCAL Context benchmarks, and the superior performance shows the proposed architecture is effective and powerful in promoting different dense prediction tasks simultaneously.

Ye Mo, Zirui Shao, Kai Ye et al.

Multimodal large language models (MLLMs) have made significant progress in document understanding. However, the information-dense nature of document images still poses challenges, as most queries depend on only a few relevant regions, with the rest being redundant. Existing one-pass MLLMs process entire document images without considering query relevance, often failing to focus on critical regions and producing unfaithful responses. Inspired by the human coarse-to-fine reading pattern, we introduce Doc-CoB (Chain-of-Box), a simple-yet-effective mechanism that integrates human-style visual reasoning into MLLM without modifying its architecture. Our method allows the model to autonomously select the set of regions (boxes) most relevant to the query, and then focus attention on them for further understanding. We first design a fully automatic pipeline, integrating a commercial MLLM with a layout analyzer, to generate 249k training samples with intermediate visual reasoning supervision. Then we incorporate two enabling tasks that improve box identification and box-query reasoning, which together enhance document understanding. Extensive experiments on seven benchmarks with four popular models show that Doc-CoB significantly improves performance, demonstrating its effectiveness and wide applicability. All code, data, and models will be released publicly.

Dengyang Jiang, Dongyang Liu, Zanyi Wang et al.

Distribution Matching Distillation (DMD) distills a pre-trained multi-step diffusion model to a few-step one to improve inference efficiency. However, the performance of the latter is often capped by the former. To circumvent this dilemma, we propose DMDR, a novel framework that combines Reinforcement Learning (RL) techniques into the distillation process. We show that for the RL of the few-step generator, the DMD loss itself is a more effective regularization compared to the traditional ones. In turn, RL can help to guide the mode coverage process in DMD more effectively. These allow us to unlock the capacity of the few-step generator by conducting distillation and RL simultaneously. Meanwhile, we design the dynamic distribution guidance and dynamic renoise sampling training strategies to improve the initial distillation process. The experiments demonstrate that DMDR can achieve leading visual quality, prompt coherence among few-step methods, and even exhibit performance that exceeds the multi-step teacher.