Research map / Image generation / Fast diffusion sampling
Photolithography Overlay Map Generation with Implicit Knowledge Distillation Diffusion Transformer
Yuan-Fu Yang, Hsiu-Hui Hsiao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add implicit knowledge distillation and gated cross-attention vs baseline DiTs | IKDDiT · MPOM, 500k iterations | FID: 14.1 -> 6.8 | Improved |
| add classifier-free guidance | IKDDiT · MPOM | FID: 6.64 -> 2.68 | Improved |
| add classifier-free guidance | IKDDiT · MPOM | sFID: 5.24 -> 6.12 | Improved |
| use gated cross-attention instead of multi-head cross-attention (ablation) | IKDDiT with UCE · MPOM ablation | FID-15k: 49.83 -> 24.66 | Improved |
| use gated cross-attention instead of adaptive normalization conditioning (ablation) | IKDDiT with UCE · MPOM ablation | FID-15k: 37.42 -> 24.66 | Improved |
| use unified contrastive embedding instead of CLIP encoder (ablation) | IKDDiT gated cross-attention · MPOM ablation | FID-15k: 30.34 -> 24.66 | Improved |
| use implicit discriminator instead of self-supervised discrimination (ablation) | IKDDiT · MPOM ablation | FID-15k: 30.39 -> 24.66 | Improved |
| mask 50% of tokens instead of none (ablation) | IKDDiT · MPOM ablation | FID-15k: 27.46 -> 24.66 | Improved |
| raise mask ratio from 50% to 70% (ablation) | IKDDiT · MPOM ablation | FID-15k: 24.66 -> 123.85 | Hurt |
| add MAE reconstruction loss (lambda1 from 0 to 0.1) (ablation) | IKDDiT · MPOM ablation, lambda2=0.01 | FID-15k: 35.07 -> 24.66 | Improved |
| add discriminator loss (lambda2 from 0 to 0.01) (ablation) | IKDDiT · MPOM ablation, lambda1=0.1 | FID-15k: 29.72 -> 24.66 | Improved |
| raise lambda1 from 0.1 to 1 (ablation) | IKDDiT · MPOM ablation, lambda2=0.01 | FID-15k: 24.66 -> 27.33 | Hurt |
Closest papers by meaning
- ILILT: Implicit Learning of Inverse Lithography Technologies (ICML 2024 · similarity 0.83)
- CoT-lized Diffusion: Let's Reinforce T2I Generation Step-by-step (NeurIPS 2025 · similarity 0.83)
- OverLayBench: A Benchmark for Layout-to-Image Generation with Dense Overlaps (NeurIPS 2025 · similarity 0.83)
- DiT4Edit: Diffusion Transformer for Image Editing (AAAI 2025 · similarity 0.83)
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.83)
- Invertible Consistency Distillation for Text-Guided Image Editing in Around 7 Steps (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
