ImageNet-FID与文生图能力负相关

重点关注

• FLAT: Feedforward Latent Triangle Splatting for Geometrically Accurate Scene Generation score 7
  入选 HF Daily Papers；HF 热度: 18 upvotes (+3)；关键词(2): lightweight, real-time
• DiffusionBench: On Holistic Evaluation of Diffusion Transformers score 9
  入选 HF Daily Papers；HF 热度: 10 upvotes (+3)；有代码实现；关键词(1): text-to-image
• ChartWalker: Benchmarking the Cross-Chart RAG Task with Hierarchical Knowledge Graphs score 7
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；有代码实现；关键词(4): retrieval-augmented, RAG, agentic, reasoning
• Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning score 6
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；关键词(3): reasoning, text-to-image, open-source

也值得关注

• MotifGen: Spatiotemporal interpolation of misaligned satellite images via multi-source generative modeling, in an application to tropical cyclones score 3
  机构: INRIA
• FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation score 3
  入选 HF Daily Papers