Weili Zeng

Keqin Chen, Zhao Zhang, Weili Zeng et al.

In human conversations, individuals can indicate relevant regions within a scene while addressing others. In turn, the other person can then respond by referring to specific regions if necessary. This natural referential ability in dialogue remains absent in current Multimodal Large Language Models (MLLMs). To fill this gap, this paper proposes an MLLM called Shikra, which can handle spatial coordinate inputs and outputs in natural language. Its architecture consists of a vision encoder, an alignment layer, and a LLM. It is designed to be straightforward and simple, without the need for extra vocabularies, position encoder, pre-/post-detection modules, or external plug-in models. All inputs and outputs are in natural language form. Referential dialogue is a superset of various vision-language (VL) tasks. Shikra can naturally handle location-related tasks like REC and PointQA, as well as conventional VL tasks such as Image Captioning and VQA. Experimental results showcase Shikra's promising performance. Furthermore, it enables numerous exciting applications, like providing mentioned objects' coordinates in chains of thoughts and comparing user-pointed regions similarities. Our code, model and dataset are accessed at https://github.com/shikras/shikra.

Weili Zeng

Energy-based models parameterize the unnormalized log-probability of data samples, but there is a lack of guidance on how to construct the "energy". In this paper, we propose a Denoising-EBM which decomposes the image energy into "semantic energy" and "texture energy". We define the "semantic energy" in the latent space of DAE to model the high-level representations, and define the pixel-level reconstruction error for denoising as "texture energy". Inspired by score-based model, our model utilizes multi-scale noisy samples for maximum-likelihood training and it outputs a vector instead of a scalar for exploring a larger set of functions during optimization. After training, the semantics are first synthesized by fast MCMC through "semantic energy", and then the pixel-level refinement of semantic image will be performed to generate perfect samples based on "texture energy". Ultimately, our model can outperform most EBMs in image generation. And we also demonstrate that Denoising-EBM has top performance among EBMs for out-of-distribution detection.

Feng Tian, Yixuan Li, Weili Zeng et al.

Diffusion posterior sampling solves inverse problems by combining a pretrained diffusion prior with measurement-consistency guidance, but it often fails to recover fine details because measurement terms are applied in a manner that is weakly coupled to the diffusion noise level. At high noise, data-consistency gradients computed from inaccurate estimates can be geometrically incongruent with the posterior geometry, inducing early-step drift, spurious high-frequency artifacts, plus sensitivity to schedules and ill-conditioned operators. To address these concerns, we propose a noise--frequency Continuation framework that constructs a continuous family of intermediate posteriors whose likelihood enforces measurement consistency only within a noise-dependent frequency band. This principle is instantiated with a stabilized posterior sampler that combines a diffusion predictor, band-limited likelihood guidance, and a multi-resolution consistency strategy that aggressively commits reliable coarse corrections while conservatively adopting high-frequency details only when they become identifiable. Across super-resolution, inpainting, and deblurring, our method achieves state-of-the-art performance and improves motion deblurring PSNR by up to 5 dB over strong baselines.

Weili Zeng, Yichao Yan, Qi Zhu et al.

Text-to-image (T2I) customization aims to create images that embody specific visual concepts delineated in textual descriptions. However, existing works still face a main challenge, concept overfitting. To tackle this challenge, we first analyze overfitting, categorizing it into concept-agnostic overfitting, which undermines non-customized concept knowledge, and concept-specific overfitting, which is confined to customize on limited modalities, i.e, backgrounds, layouts, styles. To evaluate the overfitting degree, we further introduce two metrics, i.e, Latent Fisher divergence and Wasserstein metric to measure the distribution changes of non-customized and customized concept respectively. Drawing from the analysis, we propose Infusion, a T2I customization method that enables the learning of target concepts to avoid being constrained by limited training modalities, while preserving non-customized knowledge. Remarkably, Infusion achieves this feat with remarkable efficiency, requiring a mere 11KB of trained parameters. Extensive experiments also demonstrate that our approach outperforms state-of-the-art methods in both single and multi-concept customized generation.

Weili Zeng, Ziyuan Huang, Kaixiang Ji et al.

Transformer-based models have driven significant advancements in Multimodal Large Language Models (MLLMs), yet their computational costs surge drastically when scaling resolution, training data, and model parameters. A key bottleneck stems from the proliferation of visual tokens required for fine-grained image understanding. We propose Skip-Vision, a unified framework addressing both training and inference inefficiencies in vision-language models. On top of conventional token compression approaches, our method introduces two complementary acceleration strategies. For training acceleration, we observe that Feed-Forward Network (FFN) computations on visual tokens induce marginal feature updates. This motivates our Skip-FFN strategy, which bypasses FFN layers for redundant visual tokens. For inference acceleration, we design a selective KV-cache removal mechanism that prunes the skipped key-value pairs during decoding while preserving model performance. Experimental results demonstrate that Skip-Vision reduces training time by up to 35\%, inference FLOPs by 75\%, and latency by 45\%, while achieving comparable or superior performance to existing methods. Our work provides a practical solution for scaling high-performance MLLMs with enhanced efficiency.

Weili Zeng, Yichao Yan

Flow matching has recently emerged as a powerful alternative to diffusion models, providing a continuous-time formulation for generative modeling and representation learning. Yet, we show that this framework suffers from a fundamental instability in the low-noise regime. As noise levels approach zero, arbitrarily small perturbations in the input can induce large variations in the velocity target, causing the condition number of the learning problem to diverge. This ill-conditioning not only slows optimization but also forces the encoder to reallocate its limited Jacobian capacity toward noise directions, thereby degrading semantic representations. We provide the first theoretical analysis of this phenomenon, which we term the low-noise pathology, establishing its intrinsic link to the structure of the flow matching objective. Building on these insights, we propose Local Contrastive Flow (LCF), a hybrid training protocol that replaces direct velocity regression with contrastive feature alignment at small noise levels, while retaining standard flow matching at moderate and high noise. Empirically, LCF not only improves convergence speed but also stabilizes representation quality. Our findings highlight the critical importance of addressing low-noise pathologies to unlock the full potential of flow matching for both generation and representation learning.

Tonghe Li, Jixin Liu, Weili Zeng et al.

In the context of continuously rising global air traffic, efficient and safe Conflict Detection and Resolution (CD&R) is paramount for air traffic management. Although Deep Reinforcement Learning (DRL) offers a promising pathway for CD&R automation, existing approaches commonly suffer from a "unimodal bias" in their policies. This leads to a critical lack of decision-making flexibility when confronted with complex and dynamic constraints, often resulting in "decision deadlocks." To overcome this limitation, this paper pioneers the integration of diffusion probabilistic models into the safety-critical task of CD&R, proposing a novel autonomous conflict resolution framework named Diffusion-AC. Diverging from conventional methods that converge to a single optimal solution, our framework models its policy as a reverse denoising process guided by a value function, enabling it to generate a rich, high-quality, and multimodal action distribution. This core architecture is complemented by a Density-Progressive Safety Curriculum (DPSC), a training mechanism that ensures stable and efficient learning as the agent progresses from sparse to high-density traffic environments. Extensive simulation experiments demonstrate that the proposed method significantly outperforms a suite of state-of-the-art DRL benchmarks. Most critically, in the most challenging high-density scenarios, Diffusion-AC not only maintains a high success rate of 94.1% but also reduces the incidence of Near Mid-Air Collisions (NMACs) by approximately 59% compared to the next-best-performing baseline, significantly enhancing the system's safety margin. This performance leap stems from its unique multimodal decision-making capability, which allows the agent to flexibly switch to effective alternative maneuvers.