arXiv subjects

Yuchen Bao

Publications and source records attributed to Yuchen Bao.

4 recordsLinked to original sources

arXiv2026

ReNFT: Repairing Mode Collapse in Reward Post-Training via Internal Probability-Mass Recalibration

Reward post-training of diffusion generators inevitably concentrates probability mass on a few reward-favored modes, a mode collapse that erases within-prompt diversity. Existing methods for mitigating collapse rely on external signals or interfaces, augmenting the reward with perceptual objectives, adjusting reference regularization, or modifying the text encoder, but none repairs an adapter that has already collapsed while preserving the acquired reward. We observe that online post-training primarily reallocates probability mass over capabilities inherited from pretraining rather than learning new visual content. Collapse is therefore suppression, not deletion, and can be reversed from within the generator. We propose ReNFT, which repairs a high-reward, low-diversity adapter through internal probability-mass recalibration. Unconditional probes first prioritize "anti-hub" prompts where the prompt-independent bias is easiest to expose. Two policy-dominated mixed routes then generate matched counterfactual proposals from the same prompt and initial noise, one probing the frozen base direction for suppressed alternatives and the other exposing the post-trained unconditional tendency. Reward ranking with an adaptive flipping guard assigns pull and push roles, and a joint-and-paired NFT update realizes the repair. On PickScore and GenEval, ReNFT retains 98.9% and 99.0% of NFT's reward while improving DreamSim-Div by 58.8% and 55.0%, respectively, offering a complementary alternative to external interventions.

cs.LG↗

arXiv2026

A New Pathway to Single Be Stars: Ejected Companions from Type Ia Supernovae

Be stars are rapid rotators generally produced by binary interactions. The single Be stars in the observations pose challenges to the Be star formation theory. In this paper, we propose a new pathway for the formation of single Be stars, in which the Be star is taken as the ejected companion star from a Type Ia supernova (SN Ia) explosion. Recent numerical simulations suggest that explosive oxygen burning, initiated via the convective Urca process in certain helium (He) stars near the Chandrasekhar mass limit, can set off a SN Ia. Based on this proposition, we further demonstrate that about $0.4\%$ of He star + main-sequence (MS) star binaries may evolve into single Be stars, where the MS star is spun up due to the mass accretion from the He star, and then the He star explodes as a SN Ia. We employ detailed binary evolutionary simulations and find the parameter space that would produce single Be stars via the SN Ia channel. Around $22\%$ of Be stars from the SN Ia progenitor channel exhibit peculiar tangential velocities exceeding $24\ \rm km/s$, classifying them as runaway stars. This suggests that the SN Ia channel plays a meaningful role in forming single Be stars, particularly within the runaway star population.

astro-ph.SR↗

arXiv2025

TripleFDS: Triple Feature Disentanglement and Synthesis for Scene Text Editing

Scene Text Editing (STE) aims to naturally modify text in images while preserving visual consistency, the decisive factors of which can be divided into three parts, i.e., text style, text content, and background. Previous methods have struggled with incomplete disentanglement of editable attributes, typically addressing only one aspect - such as editing text content - thus limiting controllability and visual consistency. To overcome these limitations, we propose TripleFDS, a novel framework for STE with disentangled modular attributes, and an accompanying dataset called SCB Synthesis. SCB Synthesis provides robust training data for triple feature disentanglement by utilizing the "SCB Group", a novel construct that combines three attributes per image to generate diverse, disentangled training groups. Leveraging this construct as a basic training unit, TripleFDS first disentangles triple features, ensuring semantic accuracy through inter-group contrastive regularization and reducing redundancy through intra-sample multi-feature orthogonality. In the synthesis phase, TripleFDS performs feature remapping to prevent "shortcut" phenomena during reconstruction and mitigate potential feature leakage. Trained on 125,000 SCB Groups, TripleFDS achieves state-of-the-art image fidelity (SSIM of 44.54) and text accuracy (ACC of 93.58%) on the mainstream STE benchmarks. Besides superior performance, the more flexible editing of TripleFDS supports new operations such as style replacement and background transfer. Code: https://github.com/yusenbao01/TripleFDS

cs.CV↗

arXiv2025

A Be Star + He Star Binary as an Indicator of a Binary Mass Transfer Phase

The rapid rotation of Be stars is supposed to mainly originate from binary evolution. In recent years, more and more Be stars with helium (He) star companions have been discovered, which provides a significant opportunity to study binary interaction physics. In this work, we perform binary population synthesis with an updated binary mass transfer stability criterion and try to understand the details of mass transfer processes by constructing a series of Be star + He star (BeHe) binary populations. We found that the simulations and the observations can be divided into two groups according to the masses of components, corresponding to the two distinct evolutionary processes during the mass transfer. In particular, we found that the mass ratios of BeHe binaries may be taken as a probe of the initial mass ratios of the primordial binaries. Moreover, the results suggest that a higher mass transfer efficiency ($\gtrsim 0.5$) supports the observations better. The simulations predicted too many Be star binaries experiencing Case B mass transfer, which conflicts with the observations. The reason is due to either observational selection effects or unclear physical factors.

astro-ph.SR↗