~ similar to 2606.23682· 19 results
Yuheng Chen, Teng Hu, Yuji Wang, Qingdong He +2 more
The paper proposes ST-DRC, a Spatial-Temporal Decoupled Reference Conditioning framework that effectively balances high-level semantic control and low-level identity fidelity for text-to-video generat…
Bin Lin, Zheyuan Liu, Chenguo Lin, Sixiang Chen +7 more
This paper introduces GEAR, a method for training a vector-quantized tokenizer and an autoregressive generator jointly and end-to-end, resolving the issue of non-differentiable VQ indices.
Canmiao Fu, Chen Li, Guohui Zhang, Hao Lu +12 more
WeLM-Image is a unified image foundation model for text-to-image generation, instruction-guided editing, and multi-reference synthesis.
The paper proposes VRPO, a reinforcement learning-based optimization strategy that replaces static alignment losses in diffusion models, significantly improving both convergence and image fidelity.
This paper proposes SeFi-Image, a text-to-image foundation model built upon semantic-first diffusion, and demonstrates its effectiveness on various benchmarks.
Zanyi Wang, Xin Lin, Haodong Li, Dengyang Jiang +2 more
This paper proposes ReChannel, a method for dense prediction using a pretrained DiT model, which keeps the encoder but removes the decoder and adapts it with task LoRA. ReChannel maps each token to it…
The paper introduces SynCity 3000, a framework for generating large, coherent 3D scenes using a convolutional generator, addressing the scarcity of 3D scene data for training.
This paper investigates the effect of pseudorandom number generator orbits on diffusion models' training and generation, revealing learnable input and significant impact on loss and quality.
The paper analyzes token reduction for efficient unified VLM training, finding that while task-specific acceleration saves computation, it destroys the mutual performance gains achieved through joint…
This paper introduces MFlow, an adaptive parameterization for pixel-space diffusion models that learns to interpolate between two prediction methods, and ConvRefiner, a lightweight module to improve h…
The paper proposes a disentangled representation framework to significantly improve few-shot layout-to-image generation by separating semantic identity from local visual details, thereby mitigating re…
Lu Liu, Huiyu Duan, Chenxin Zhu, Jintong Lu +5 more
The paper introduces LL-Bench, a comprehensive benchmark for evaluating large-scale generative models on low-level vision tasks, and proposes LL-Score, an MLLM-based evaluator that better aligns quali…
The paper proposes Distribution-Aligned Self-Distillation (DASD) to improve self-distillation by dynamically filtering high-perplexity tokens, thereby preserving useful logical knowledge while suppres…