✨Condition Errors Refinement in Autoregressive Image Generation with Diffusion Loss

📝 Summary:
This study refines autoregressive image generation with diffusion loss, showing patch denoising effectively mitigates condition errors. A novel Optimal Transport based condition refinement method is introduced to ensure convergence to an ideal condition distribution, outperforming prior methods.

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.07022
• PDF: https://arxiv.org/pdf/2602.07022

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ImageGeneration #DiffusionModels #AutoregressiveModels #OptimalTransport #MachineLearning

✨Unified Latents (UL): How to train your latents

📝 Summary:
Unified Latents UL learns joint latent representations using diffusion prior regularization and decoding. It achieves competitive FID of 1.4 on ImageNet-512 with fewer training FLOPs and sets a new state of the art FVD of 1.3 on Kinetics-600.

🔹 Publication Date: Published on Feb 19

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.17270
• PDF: https://arxiv.org/pdf/2602.17270

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#GenerativeAI #DiffusionModels #LatentSpace #ImageGeneration #VideoGeneration

✨Accelerating Masked Image Generation by Learning Latent Controlled Dynamics

📝 Summary:
MIGM-Shortcut accelerates masked image generation by learning a lightweight model to predict feature evolution velocity from previous features and sampled tokens. This achieves over 4x speedup with maintained quality on state-of-the-art models.

🔹 Publication Date: Published on Feb 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23996
• PDF: https://arxiv.org/pdf/2602.23996
• Github: https://github.com/Kaiwen-Zhu/MIGM-Shortcut

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ImageGeneration #DeepLearning #GenerativeAI #ComputerVision #AI

✨Enhancing Spatial Understanding in Image Generation via Reward Modeling

📝 Summary:
Text-to-image models struggle with complex spatial relationships. This paper introduces SpatialScore, a reward model trained on 80k preference pairs, to evaluate and improve spatial accuracy. It significantly enhances spatial understanding in image generation via reinforcement learning.

🔹 Publication Date: Published on Feb 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.24233
• PDF: https://arxiv.org/pdf/2602.24233
• Project Page: https://dagroup-pku.github.io/SpatialT2I/
• Github: https://github.com/DAGroup-PKU/SpatialT2I

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ImageGeneration #TextToImage #SpatialAI #RewardModeling #DeepLearning

✨AutoFigure-Edit: Generating Editable Scientific Illustration

📝 Summary:
AutoFigure-Edit generates editable scientific illustrations from text and reference images. It improves editability, style control, and efficiency by combining long-context understanding and native SVG editing for high-quality, flexible refinement.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2603.06674
• PDF: https://arxiv.org/pdf/2603.06674
• Project Page: https://deepscientist.cc/
• Github: https://github.com/ResearAI/AutoFigure-Edit

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#AI #ScientificIllustration #ImageGeneration #SVG #DeepLearning

✨SNCE: Geometry-Aware Supervision for Scalable Discrete Image Generation

📝 Summary:
SNCE is a novel training objective for large-codebook discrete image generators. It supervises models with a soft categorical distribution over neighboring tokens, based on embedding proximity, instead of hard one-hot targets. This approach significantly improves convergence speed and overall gen...

🔹 Publication Date: Published on Mar 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.15150
• PDF: https://arxiv.org/pdf/2603.15150

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ImageGeneration #DeepLearning #ComputerVision #GeometryAware #AIResearch

✨GenMask: Adapting DiT for Segmentation via Direct Mask

📝 Summary:
GenMask directly trains a DiT for joint image generation and segmentation using a novel timestep sampling strategy. This strategy emphasizes extreme noise for masks, enabling harmonious training. It outperforms indirect adaptation, simplifying the workflow.

🔹 Publication Date: Published on Mar 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.23906
• PDF: https://arxiv.org/pdf/2603.23906

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#Segmentation #ImageGeneration #DiT #DeepLearning #ComputerVision

✨Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning

📝 Summary:
This paper introduces process-driven image generation, an iterative method with interleaved textual and visual reasoning. It decomposes synthesis into planning, drafting, reflecting, and refining steps. Dense step-wise supervision ensures consistency and interpretability of intermediate states.

🔹 Publication Date: Published on Apr 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04746
• PDF: https://arxiv.org/pdf/2604.04746

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#ImageGeneration #GenerativeAI #ArtificialIntelligence #DeepLearning #ComputerVision

🔥 D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models

💡 The paper introduces D-OPSD, a new training approach for diffusion models that enables efficient supervised fine-tuning while preserving few-step inference capabilities. The current landscape of high-performance image generation models is shifting from inefficient multi-step models to efficient few-step models, but these models are challenging to fine-tune using traditional techniques. The problem with traditional fine-tuning methods is that they compromise the model's inherent few-step inference capability.

To address this issue, the authors propose D-OPSD, which leverages on-policy self-distillation with text and multimodal features. The method works by making the model act as both the teacher and the student, where the student is conditioned only on the text feature, and the teacher is conditioned on the multimodal feature of both the text prompt and the target image. The training process minimizes the difference between the predicted distributions over the student's own roll-outs, allowing the model to learn new concepts and styles without sacrificing its original few-step capacity.

The key contribution of D-OPSD is that it enables on-policy learning during supervised fine-tuning, which allows the model to learn from its own trajectory and under its own supervision. This approach enables the model to inherit the in-context capabilities of its encoder, making it possible to fine-tune the model continuously without compromising its few-step inference capability. The results show that D-OPSD enables efficient supervised fine-tuning for diffusion models, making it a promising approach for high-performance image generation models.

📅 Published on May 6

🔗 Links:
• arXiv: https://arxiv.org/abs/2605.05204
• PDF: https://arxiv.org/pdf/2605.05204
• Project Page: https://vvvvvjdy.github.io/d-opsd/
• GitHub: https://github.com/vvvvvjdy/D-OPSD ⭐ 24

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#DiffusionModels #SelfDistillation #FewShotLearning #ImageGeneration #MultimodalLearning

🔥 GenEvolve: Self-Evolving Image Generation Agents via Tool-Orchestrated Visual Experience Distillation

💡 The paper proposes a self-evolving image generation framework called GenEvolve that improves generative capabilities through iterative learning and reference-based prompting. The problem addressed is that high-quality image generation often requires combining a model's internal generative ability with external resources, and existing methods have limitations in handling diverse and demanding requests.

The GenEvolve framework models each generation attempt as a tool-orchestrated trajectory, where the agent gathers evidence, selects references, invokes generation skills, and composes them into a prompt-reference program. Unlike existing methods that rely on image-level scalar rewards, GenEvolve compares multiple trajectories for the same request and abstracts best-worst differences into structured visual experience.

This visual experience is provided to a privileged teacher branch, which uses visual experience distillation to provide dense token-level supervision to a student branch. This helps the student internalize better search, knowledge activation, reference selection, and prompt construction. The authors also construct GenEvolve-Data and GenEvolve-Bench to evaluate the framework.

The results show that GenEvolve achieves substantial gains over strong baselines, achieving state-of-the-art performance among current image-generation frameworks. The experiments on public benchmarks and GenEvolve-Bench demonstrate the effectiveness of the proposed framework. Overall, the paper contributes a novel self-evolving image generation framework that can effectively handle diverse and demanding generation challenges.

📅 Published on May 20

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2605.21605
• PDF: https://arxiv.org/pdf/2605.21605
• Project Page: https://ephemeral182.github.io/GenEvolve/

🤖 Models citing this paper:
• https://huggingface.co/MeiGen-AI/GenEvolve

📊 Datasets citing this paper:
• https://huggingface.co/datasets/MeiGen-AI/GenEvolve-Data-Bench

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#ComputerVision #ImageGeneration #GenerativeModels #SelfEvolvingSystems #DeepLearning