arXiv subjects

Pipei Huang

Publications and source records attributed to Pipei Huang.

17 recordsLinked to original sources

arXiv2026

Harnessing Intrinsic Subject-Aware Attention for Controllable Multi-Subject Video Generation

Multi-subject video generation faces two key challenges: uncontrollable fidelity strength and potential semantic drift. We address these by analyzing the internal mechanisms of Diffusion Transformers (DiTs). We found that certain attention blocks naturally form an Intrinsic Spatial Grounding Map (ISGM) that precisely locates reference subjects. Building on this insight, we propose Dual-phase Intrinsic Attention Leveraging (DIAL), a framework that uses these internal signals for both training and inference. In low-noise stages, we use ISGM to guide the attention mechanism, allowing precise control over fidelity strength during inference without retraining. In high-noise stages, we use these same maps to automatically build preference pairs at no additional cost for Reinforcement Learning (RL). This RL procedure effectively anchors the model's attention to reference subjects and mitigates semantic drift. Extensive experiments show that DIAL significantly outperforms baseline models on the OpenS2V-Eval benchmark, consistently improving identity consistency and enabling controllable fidelity strength.

cs.CV↗

arXiv2026

RAGDiffusion++: From Macro-Retrieval to Micro-Fidelity Alignment for Garment Generation

Standard clothing asset generation---restoring forward-facing flat-lay garment images from diverse real-world contexts---holds immense commercial value yet demands both macroscopic topological accuracy and microscopic physical fidelity. Although our previous work RAGDiffusion effectively eradicated large-scale structural hallucinations via retrieval-augmented macro-constraints, achieving industrial-grade micro-texture realism remains an unsolved bottleneck. We formally identify this limitation as High-Frequency Trajectory Collapse: supervised fine-tuning (SFT) converges to the conditional mean of the training distribution, which is dominated by smooth, low-frequency textures, causing high-frequency patterns (e.g., fabric weaves, intricate logos) to become nearly un-sampleable. Naively applying Reinforcement Learning (RL) post-training further triggers Artifact Hacking, where models exploit semantic biases in generic reward models by generating deceptive checkerboard noise. Our key insight is that RL can fundamentally reshape the sampling distribution of flow models---elevating the probability of high-fidelity trajectories under accurate reward guidance---while adversarial regularization prevents exploitation of reward blind spots. Realizing this principle requires three prerequisites: (i)inherent capacity, established through a 27,725-pair high-complexity garment dataset (STGarment-Plus) and a Dual-Image-Stream FLUX architecture upgrade; (ii)perceptive reward, provided by a novel attribute-aware reward model (Garment-RM) trained on 500K images via fine-grained contrastive learning, achieving 84.67% human preference accuracy; and (iii)hacking prevention, enforced by our Adversarial-Regularized GRPO (AR-GRPO) strategy that integrates a dynamic discriminator into the RL sampling trajectory to penalize artifacts while enriching authentic high-frequency details.

cs.CV↗

arXiv2026

Self-OPD: On-Policy Distillation for Flow Matching Models without Teacher

On-policy distillation (OPD), which leverages a pre-trained, specialized teacher model to provide dense supervisory signals, has achieved significant success in Large Language Models (LLMs) and has recently been adapted to flow matching models. However, this paradigm suffers from two major issues: First, training a separate, task-specific teacher for every new objective incurs high computational costs. Second, the discrepancy between teacher and student distributions often leads to compounding errors along the generation trajectory. In this paper, we introduce \textbf{Self-OPD}, a teacher-free OPD framework for flow matching models that turns the student's own self-exploration into step-wise supervision. At each timestep, Self-OPD branches the deterministic next-state prediction into $K$ stochastic SDE candidates, rolls them out with the ODE sampler, and compares their rewards against a deterministic self-reference baseline to obtain normalized advantages. The velocity field is optimized with an all-branch pull-push objective, where high-advantage branches attract the student and low-advantage branches repel it under direction-aware attenuation and SDE-variance normalization. For multi-objective alignment, Self-OPD fuses normalized scores at the reward level, avoiding direct gradient conflict. Experiments on single and mixed reward benchmarks show that Self-OPD outperforms prior RL and OPD methods without task-specific teachers.

cs.CV↗

arXiv2026

InfinityEdit: Infinite Video Editing with a Lightweight Edit-Ignition Adapter

With large pretrained models, existing methods have effectively improved instruction-based video editing. However, most of them rely on an in-place editing assumption. They align the edited video with the given source clip frame by frame over a fixed time span. This pattern fails for open-ended streams, e.g., restyling a live game or applying a camera move to an ongoing shot. In such cases, edits must extend to future frames as they arrive, rather than be applied to a static input clip. In this paper, we study this setting and name it infinite video editing: given a preceding segment and an edit request, a model must generate the next segment that continues the stream while applying the requested edit. This process repeats as an unbounded sequence of edit instructions arrives. This task brings two challenges: the edit must be a faithful continuation rather than a frame-wise rewrite, and generation quality must remain stable as edits accumulate. To address them, we first design a data-collection pipeline for infinite video editing. Based on the collected data, we propose InfinityEdit, a lightweight edit adapter that equips a streaming video generator with unbounded editing ability. The adapter contains three attention modules. History cross-attention guides the denoising frames using the input frames. Temporal causal self-attention keeps temporal cues flowing only from earlier frames to later ones. Edit cross-attention injects the edit request into generation. During inference, the adapter is activated only in the chunk where an edit request arrives. Subsequent chunks are generated by the original model with a reset anchor frame. This scheme applies the edit while preserving the original model's infinite generation ability. Extensive experiments show that InfinityEdit faithfully continues the stream under each edit, and stays stable over unbounded edit sequences.

cs.CV↗

arXiv2026

RL-Native Distillation: Exploiting Scored Trajectories for Few-Step Image Generation

Efficient text-to-image generation requires both reinforcement-learning (RL)-based reward alignment and few-step distillation, yet these procedures are typically performed sequentially, increasing training cost and risking the loss of reward gains during compression. We instead take an RL-native perspective: diffusion RL already generates reward-scored finite-step trajectories, whose intermediate states provide a natural source of distillation supervision rather than a disposable byproduct of sampling. Based on this insight, we propose REST (Reward-Enhanced Scored-Trajectory Distillation), a single-stage RL-distillation co-training framework that attaches a decoupled student to an arbitrary RL teacher. The student learns segment-wise from the teacher's evolving rollout trajectories while leaving the original teacher optimization unchanged. To prevent uniform imitation from preserving undesirable low-reward behaviors, we further introduce Advantage-Modulated Distillation (AMD), which transforms rollout advantages into signed weights over a base distillation loss. AMD strengthens supervision from preferred trajectories and mildly repels the student from low-reward ones. The resulting framework is lightweight and plug-and-play, requires no extra image rollouts, no separate distillation dataset, and no adversarial training. Experiments on compositional generation, visual text rendering, and human-preference alignment show that REST enables few-step CFG-free inference that matches or surpasses its 40-step RL teacher, with an overall additional training cost below 25% over pure RL. REST improves DrawBench PickScore over RTDMD by 0.82 while requiring only one-fifth of the training iterations.

cs.CV↗

arXiv2026

LoomVideo: Unifying Multimodal Inputs into Video Generation and Editing

Developing unified video generation and editing models capable of interpreting interleaved multimodal inputs is a promising yet challenging frontier field. Existing unified frameworks predominantly rely on massive models (typically 13B parameters or more) and incorporate source video conditions for editing by concatenating sequence tokens. This concatenation inevitably doubles the sequence length, quadrupling the computational complexity of the self-attention mechanism and introducing prohibitive overhead. To address these bottlenecks, we present LoomVideo, a highly efficient 5B-parameter unified architecture for both video generation and editing. LoomVideo replaces the standard text encoder with a Multimodal Large Language Model (MLLM) and employs Deepstack injection mechanism to align multi-layer MLLM features with the Diffusion Transformer (DiT). Crucially, we introduce a zero-overhead Scale-and-Add conditioning approach for video editing. By scaling and directly adding the clean source video latent to the noised target latent, this elegant design eliminates the need for token concatenation, drastically reducing computational cost while maintaining robust capabilities for complex, non-rigid edits. Furthermore, a Negative Temporal RoPE strategy is seamlessly integrated to handle multiple reference images. Extensive experiments demonstrate that our compact 5B model achieves state-of-the-art or highly competitive performance across comprehensive benchmarks, exhibiting exceptional superiority in e-commerce and fashion generation scenarios. Benefiting from the zero-overhead conditioning mechanism, LoomVideo achieves at least a 5.41x acceleration in inference speed compared to models of similar capabilities, paving the way for highly practical and efficient video foundation models.

cs.CV↗

arXiv2026

DynFrame: Adaptive Reasoning-Driven Multimodal Framework with Dynamic Frame Augmentation for Complex Video Understanding

Recent video multimodal large language models (MLLMs) increasingly couple step-by-step reasoning with on-demand visual evidence retrieval, allowing models to revisit relevant video segments during inference. However, two structural gaps remain in existing thinking-with-video systems. (i) Sampling density is not a learnable decision: existing methods may let the model decide where to look, but the per-window frame rate is largely fixed. As a result, fine-grained evidence is often recovered through repeated retrieval calls, which increases inference context length and training difficulty. (ii) Retrieval and answer generation are usually optimized with a single trajectory-level advantage, so the "where to look" tokens and the "how to answer" tokens receive the same credit even when one is correct and the other is not. To address these gaps, we present DynFrame, a framework that emits the temporal window and the sampling density as native tokens within a single autoregressive pass. This learnable span-density retrieval enables acquiring multi-granularity evidence with a single retrieval step. Based on the above tokenized retrieval interface, we further introduce Segment-Decoupled GRPO (SD-GRPO), which splits each rollout at the retrieval boundary and assigns role-specific token-level advantages, separately crediting the sampling decision and the answer. Trained on the curated DM-CoT-74k and DM-RL-45k, DynFrame-4B is competitive with strong 7B-8B baselines across six benchmarks (NExT-GQA, Charades-STA, ActivityNet-MR, Video-MME, MLVU, LVBench), and DynFrame-8B sets new state-of-the-art on most metrics. Code is available at https://github.com/zhangguanghao523/DynFrame.

cs.CV↗

arXiv2026

Think When Needed: Adaptive Reasoning-Driven Multimodal Embeddings with a Dual-LoRA Architecture

Multimodal large language models (MLLMs) have emerged as a powerful backbone for multimodal embeddings. Recent methods introduce chain-of-thought (CoT) reasoning into the embedding pipeline to improve retrieval quality, but remain costly in both model size and inference cost. They typically employ separate reasoner and embedder with substantial parameter overhead, and generate CoT indiscriminately for every input. However, we observe that for simple inputs, discriminative embeddings already perform well, and redundant reasoning can even mislead the model, degrading performance. To address these limitations, we propose Think When Needed (TWN), a unified multimodal embedding framework with adaptive reasoning. TWN introduces a dual-LoRA architecture that attaches reasoning and embedding adapters to a shared frozen backbone, detaching gradients at their interface to mitigate gradient conflicts introduced by joint optimization while keeping parameters close to a single model. Building on this, an adaptive think mechanism uses a self-supervised routing gate to decide per input whether to generate CoT, skipping unnecessary reasoning to reduce inference overhead and even improve retrieval quality. We further explore embedding-guided RL to optimize CoT quality beyond supervised training. On the 78 tasks of MMEB-V2, TWN achieves state-of-the-art embedding quality while being substantially more efficient than existing generative methods, requiring only 3-5% additional parameters relative to the backbone and up to 50% fewer reasoning tokens compared to the full generative mode.

cs.CV↗

arXiv2026

simpleposter: A simple baseline for product poster generation

Product poster generation poses distinct challenges beyond general poster design, requiring both faithful preservation of product appearance and precise control over dense, multi-line text layouts. Prior methods typically adopt inpainting frameworks augmented with auxiliary modules such as ControlNet and OCR encoders. However, these approaches introduce architectural complexity and computational overhead while still suffering from text errors and subject extension artifacts. We present SimplePoster, a simple yet effective inpainting-based framework that achieves faithful subject preservation and accurate, position-controllable text rendering without external controllers. Our approach builds on two observations: (1) full-parameter fine-tuning of the base model effectively suppresses subject extension, outperforming ControlNet-based alternatives; and (2) a zero-cost character-level position encoding enables geometry-aware text generation without dedicated layout modules. Experiments show that SimplePoster achieves a $98.7\%$ subject preservation rate, compared to $55.2\%$ for SeedEdit 3.0 and $85.3\%$ for PosterMaker, while also improving text rendering accuracy. Code, models, benchmark and a part of training data will be available at https://github.com/Alibaba-YuFeng/SIMPLEPOSTER

cs.CV↗

arXiv2026

Diffusion-APO: Trajectory-Aware Direct Preference Alignment for Video Diffusion Transformers

Efficiently aligning large-scale video diffusion models with human intent requires a scalable and trajectory-aware pathway that bridges the inherent discrepancy between training noise distributions and practical inference trajectories. While existing paradigms such as Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO) attempt to address this, they are often hindered by either reliance on bias-prone, complex reward models or suboptimal timestep sampling. In this paper, we propose Diffusion-APO (Aligned Preference Optimization), a trajectory-aware algorithm that resolves this misalignment by synchronizing training noise with inference-time denoising paths to maximize gradient signal efficacy. To translate this algorithmic innovation into a practical solution, we introduce a unified and modular RLHF framework that integrates online ranking, half-online anchoring, offline refinement, and distillation-aware drift correction. This framework enables flexible, multi-stage preference alignment across diverse data and computational constraints without relying on scalar-reward-based policy gradients. Through extensive experiments, we demonstrate that Diffusion-APO consistently outperforms standard baselines in visual quality and instruction following, while effectively preserving generative fidelity during model acceleration, providing a robust, end-to-end pathway for scalable video diffusion alignment.

cs.CV↗

arXiv2026

PromptEcho: Annotation-Free Reward from Vision-Language Models for Text-to-Image Reinforcement Learning

Reinforcement learning (RL) can improve the prompt following capability of text-to-image (T2I) models, yet obtaining high-quality reward signals remains challenging: CLIP Score is too coarse-grained, while VLM-based reward models (e.g., RewardDance) require costly human-annotated preference data and additional fine-tuning. We propose PromptEcho, a reward construction method that requires \emph{no} annotation and \emph{no} reward model training. Given a generated image and a guiding query, PromptEcho computes the token-level cross-entropy loss of a frozen VLM with the original prompt as the label, directly extracting the image-text alignment knowledge encoded during VLM pretraining. The reward is deterministic, computationally efficient, and improves automatically as stronger open-source VLMs become available. For evaluation, we develop DenseAlignBench, a benchmark of concept-rich dense captions for rigorously testing prompt following capability. Experimental results on two state-of-the-art T2I models (Z-Image and QwenImage-2512) demonstrate that PromptEcho achieves substantial improvements on DenseAlignBench (+26.8pp / +16.2pp net win rate), along with consistent gains on GenEval, DPG-Bench, and TIIFBench without any task-specific training. Ablation studies confirm that PromptEcho comprehensively outperforms inference-based scoring with the same VLM, and that reward quality scales with VLM size. We will open-source the trained models and the DenseAlignBench.

cs.CV↗

arXiv2026

Denoising as Path Planning: Training-Free Acceleration of Diffusion Models with DPCache

Diffusion models have demonstrated remarkable success in image and video generation, yet their practical deployment remains hindered by the substantial computational overhead of multi-step iterative sampling. Among acceleration strategies, caching-based methods offer a training-free and effective solution by reusing or predicting features across timesteps. However, existing approaches rely on fixed or locally adaptive schedules without considering the global structure of the denoising trajectory, often leading to error accumulation and visual artifacts. To overcome this limitation, we propose DPCache, a novel training-free acceleration framework that formulates diffusion sampling acceleration as a global path planning problem. DPCache constructs a Path-Aware Cost Tensor from a small calibration set to quantify the path-dependent error of skipping timesteps conditioned on the preceding key timestep. Leveraging this tensor, DPCache employs dynamic programming to select an optimal sequence of key timesteps that minimizes the total path cost while preserving trajectory fidelity. During inference, the model performs full computations only at these key timesteps, while intermediate outputs are efficiently predicted using cached features. Extensive experiments on DiT, FLUX, and HunyuanVideo demonstrate that DPCache achieves strong acceleration with minimal quality loss, outperforming prior acceleration methods by $+$0.031 ImageReward at 4.87$\times$ speedup and even surpassing the full-step baseline by $+$0.028 ImageReward at 3.54$\times$ speedup on FLUX, validating the effectiveness of our path-aware global scheduling framework. Code is available at https://github.com/argsss/DPCache.

cs.CV↗

arXiv2019

POG: Personalized Outfit Generation for Fashion Recommendation at Alibaba iFashion

Increasing demand for fashion recommendation raises a lot of challenges for online shopping platforms and fashion communities. In particular, there exist two requirements for fashion outfit recommendation: the Compatibility of the generated fashion outfits, and the Personalization in the recommendation process. In this paper, we demonstrate these two requirements can be satisfied via building a bridge between outfit generation and recommendation. Through large data analysis, we observe that people have similar tastes in individual items and outfits. Therefore, we propose a Personalized Outfit Generation (POG) model, which connects user preferences regarding individual items and outfits with Transformer architecture. Extensive offline and online experiments provide strong quantitative evidence that our method outperforms alternative methods regarding both compatibility and personalization metrics. Furthermore, we deploy POG on a platform named Dida in Alibaba to generate personalized outfits for the users of the online application iFashion. This work represents a first step towards an industrial-scale fashion outfit generation and recommendation solution, which goes beyond generating outfits based on explicit queries, or merely recommending from existing outfit pools. As part of this work, we release a large-scale dataset consisting of 1.01 million outfits with rich context information, and 0.28 billion user click actions from 3.57 million users. To the best of our knowledge, this dataset is the largest, publicly available, fashion related dataset, and the first to provide user behaviors relating to both outfits and fashion items.

cs.IR↗

arXiv2019

Behavior Sequence Transformer for E-commerce Recommendation in Alibaba

Deep learning based methods have been widely used in industrial recommendation systems (RSs). Previous works adopt an Embedding&MLP paradigm: raw features are embedded into low-dimensional vectors, which are then fed on to MLP for final recommendations. However, most of these works just concatenate different features, ignoring the sequential nature of users' behaviors. In this paper, we propose to use the powerful Transformer model to capture the sequential signals underlying users' behavior sequences for recommendation in Alibaba. Experimental results demonstrate the superiority of the proposed model, which is then deployed online at Taobao and obtain significant improvements in online Click-Through-Rate (CTR) comparing to two baselines.

cs.IR↗

arXiv2019

Multi-Interest Network with Dynamic Routing for Recommendation at Tmall

Industrial recommender systems usually consist of the matching stage and the ranking stage, in order to handle the billion-scale of users and items. The matching stage retrieves candidate items relevant to user interests, while the ranking stage sorts candidate items by user interests. Thus, the most critical ability is to model and represent user interests for either stage. Most of the existing deep learning-based models represent one user as a single vector which is insufficient to capture the varying nature of user's interests. In this paper, we approach this problem from a different view, to represent one user with multiple vectors encoding the different aspects of the user's interests. We propose the Multi-Interest Network with Dynamic routing (MIND) for dealing with user's diverse interests in the matching stage. Specifically, we design a multi-interest extractor layer based on capsule routing mechanism, which is applicable for clustering historical behaviors and extracting diverse interests. Furthermore, we develop a technique named label-aware attention to help learn a user representation with multiple vectors. Through extensive experiments on several public benchmarks and one large-scale industrial dataset from Tmall, we demonstrate that MIND can achieve superior performance than state-of-the-art methods for recommendation. Currently, MIND has been deployed for handling major online traffic at the homepage on Mobile Tmall App.

cs.IR↗

arXiv2018

Multi-Level Deep Cascade Trees for Conversion Rate Prediction in Recommendation System

Developing effective and efficient recommendation methods is very challenging for modern e-commerce platforms. Generally speaking, two essential modules named "Click-Through Rate Prediction" (\textit{CTR}) and "Conversion Rate Prediction" (\textit{CVR}) are included, where \textit{CVR} module is a crucial factor that affects the final purchasing volume directly. However, it is indeed very challenging due to its sparseness nature. In this paper, we tackle this problem by proposing multi-Level Deep Cascade Trees (\textit{ldcTree}), which is a novel decision tree ensemble approach. It leverages deep cascade structures by stacking Gradient Boosting Decision Trees (\textit{GBDT}) to effectively learn feature representation. In addition, we propose to utilize the cross-entropy in each tree of the preceding \textit{GBDT} as the input feature representation for next level \textit{GBDT}, which has a clear explanation, i.e., a traversal from root to leaf nodes in the next level \textit{GBDT} corresponds to the combination of certain traversals in the preceding \textit{GBDT}. The deep cascade structure and the combination rule enable the proposed \textit{ldcTree} to have a stronger distributed feature representation ability. Moreover, inspired by ensemble learning, we propose an Ensemble \textit{ldcTree} (\textit{E-ldcTree}) to encourage the model's diversity and enhance the representation ability further. Finally, we propose an improved Feature learning method based on \textit{EldcTree} (\textit{F-EldcTree}) for taking adequate use of weak and strong correlation features identified by pre-trained \textit{GBDT} models. Experimental results on off-line data set and online deployment demonstrate the effectiveness of the proposed methods.

cs.LG↗

arXiv2018

Billion-scale Commodity Embedding for E-commerce Recommendation in Alibaba

Recommender systems (RSs) have been the most important technology for increasing the business in Taobao, the largest online consumer-to-consumer (C2C) platform in China. The billion-scale data in Taobao creates three major challenges to Taobao's RS: scalability, sparsity and cold start. In this paper, we present our technical solutions to address these three challenges. The methods are based on the graph embedding framework. We first construct an item graph from users' behavior history. Each item is then represented as a vector using graph embedding. The item embeddings are employed to compute pairwise similarities between all items, which are then used in the recommendation process. To alleviate the sparsity and cold start problems, side information is incorporated into the embedding framework. We propose two aggregation methods to integrate the embeddings of items and the corresponding side information. Experimental results from offline experiments show that methods incorporating side information are superior to those that do not. Further, we describe the platform upon which the embedding methods are deployed and the workflow to process the billion-scale data in Taobao. Using online A/B test, we show that the online Click-Through-Rate (CTRs) are improved comparing to the previous recommendation methods widely used in Taobao, further demonstrating the effectiveness and feasibility of our proposed methods in Taobao's live production environment.

cs.IR↗