Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2511.20211 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913068568543232 |
|---|---|
| author | Yu, Hao Wang, Jinglin Zhan, Jiabo Chen, Rui Wang, Zile Zhang, Huaisong Li, Hongyu Chen, Xinrui Wei, Yongxian Yuan, Chun |
| author_facet | Yu, Hao Wang, Jinglin Zhan, Jiabo Chen, Rui Wang, Zile Zhang, Huaisong Li, Hongyu Chen, Xinrui Wei, Yongxian Yuan, Chun |
| contents | Transparency-aware generation requires modeling not only RGB appearance but also alpha-based opacity and cross-layer composition, which are essential for tasks such as image matting, object removal, layer decomposition, and multi-layer content creation. However, existing RGBA-related methods remain largely fragmented, with separate pipelines designed for individual tasks. While a unified model is desirable, supervised fine-tuning alone is insufficient, as localized regression objectives cannot directly optimize the compositional fidelity, alpha-boundary precision, and structural consistency required for high-quality RGBA generation. To address this, we propose OmniAlpha, a unified multi-task reinforcement learning framework for transparency-aware generation and manipulation. OmniAlpha combines an end-to-end alpha-aware VAE and a sequence-to-sequence Diffusion Transformer, with a bi-directional layer axis in positional encoding to jointly model multiple RGBA inputs and outputs within a single forward pass. Built on a multi-task SFT cold start, it further performs GRPO-style post-training with layer-aware rewards defined on decoded RGBA outputs, enabling direct optimization of cross-layer coherence and fine transparency details. Experiments across five categories of transparency-aware tasks show that OmniAlpha consistently outperforms its unified SFT baseline and achieves strong performance against specialized expert models, including a 9.07% relative reduction in RGB L1 on layer decomposition and 74%/68% improvements over conventional matting tools on SAD/Grad for automatic matting. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_20211 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | OmniAlpha: Aligning Transparency-Aware Generation via Multi-Task Unified Reinforcement Learning Yu, Hao Wang, Jinglin Zhan, Jiabo Chen, Rui Wang, Zile Zhang, Huaisong Li, Hongyu Chen, Xinrui Wei, Yongxian Yuan, Chun Computer Vision and Pattern Recognition Artificial Intelligence Transparency-aware generation requires modeling not only RGB appearance but also alpha-based opacity and cross-layer composition, which are essential for tasks such as image matting, object removal, layer decomposition, and multi-layer content creation. However, existing RGBA-related methods remain largely fragmented, with separate pipelines designed for individual tasks. While a unified model is desirable, supervised fine-tuning alone is insufficient, as localized regression objectives cannot directly optimize the compositional fidelity, alpha-boundary precision, and structural consistency required for high-quality RGBA generation. To address this, we propose OmniAlpha, a unified multi-task reinforcement learning framework for transparency-aware generation and manipulation. OmniAlpha combines an end-to-end alpha-aware VAE and a sequence-to-sequence Diffusion Transformer, with a bi-directional layer axis in positional encoding to jointly model multiple RGBA inputs and outputs within a single forward pass. Built on a multi-task SFT cold start, it further performs GRPO-style post-training with layer-aware rewards defined on decoded RGBA outputs, enabling direct optimization of cross-layer coherence and fine transparency details. Experiments across five categories of transparency-aware tasks show that OmniAlpha consistently outperforms its unified SFT baseline and achieves strong performance against specialized expert models, including a 9.07% relative reduction in RGB L1 on layer decomposition and 74%/68% improvements over conventional matting tools on SAD/Grad for automatic matting. |
| title | OmniAlpha: Aligning Transparency-Aware Generation via Multi-Task Unified Reinforcement Learning |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2511.20211 |