Research map / Image generation / Image restoration
Trans-Adapter: A Plug-and-Play Framework for Transparent Image Inpainting
Yuekun Dai, Haitian Li, Shangchen Zhou, Chen Change Loy
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 spatial alignment module (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AS: 5.685 -> 6.097 | Improved |
| add spatial alignment module (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AEQ: 0.9897 -> 0.9878 | Hurt |
| add cross-domain self-attention (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AS: 6.065 -> 6.097 | Improved |
| use designed modules instead of original AnimateDiff (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | LPIPS: 0.0459 -> 0.0408 | Improved |
| add MAGICK training data (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | LPIPS: 0.0435 -> 0.0408 | Improved |
| add own 35K PNG-stock dataset (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | LPIPS: 0.0457 -> 0.0408 | Improved |
| use LayerDiffuse RGB padding instead of grey background (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AEQ: 0.9717 -> 0.9878 | Improved |
| use LayerDiffuse RGB padding instead of Telea padding (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AEQ: 0.9853 -> 0.9878 | Improved |
| add alpha-map LoRA stage (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench pure noise | LPIPS: 0.0769 -> 0.0591 | Improved |
| unfreeze LoRA in stage 2 (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AEQ: 0.9831 -> 0.9878 | Improved |
| train LoRA on both RGB and alpha instead of alpha only (ablation) | Trans-Adapter (SD1.5-inpaint) · LayerBench blended noise | AEQ: 0.9863 -> 0.9878 | Improved |
| joint RGBA inpainting instead of inpaint then BiRefNet matting | SD1.5-inpaint · LayerBench blended noise | LPIPS: 0.0515 -> 0.0408 | Improved |
Closest papers by meaning
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.85)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.85)
- TransPixeler: Advancing Text-to-Video Generation with Transparency (CVPR 2025 · similarity 0.85)
- InstantSticker: Realistic Decal Blending via Disentangled Object Reconstruction (AAAI 2025 · similarity 0.85)
- DesignEdit: Unify Spatial-Aware Image Editing via Training-free Inpainting with a Multi-Layered Latent Diffusion Framework (AAAI 2025 · similarity 0.84)
- DRIP: Unleashing Diffusion Priors for Joint Foreground and Alpha Prediction in Image Matting (NeurIPS 2024 · similarity 0.84)
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.
