Research map / Image generation / Autoregressive and latent generation
Transformer Based Pluralistic Image Completion With Reduced Information Loss
Qiankun Liu, Yuqi Jiang, Zhentao Tan, Dongdong Chen, Ying Fu, Qi Chu, Gang Hua, Nenghai Yu
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 patch-based encoder (vs CNN encoder) (ablation) | PUT conference config · Places2 256x256 (mask 10-60%) | FID: 179.294 -> 22.121 | Improved |
| add dual-codebook (ablation) | PUT conference config · Places2 256x256 (mask 10-60%) | FID: 28.634 -> 22.121 | Improved |
| add dual-codebook (ablation) | PUT conference config · FFHQ 256x256 (mask 10-60%) | FID: 13.960 -> 14.554 | Hurt |
| add reference-branch guidance in decoder (ablation) | PUT conference config · Places2 256x256 (mask 10-60%) | FID: 25.084 -> 22.121 | Improved |
| feed un-quantized features to UQ-Transformer (ablation) | PUT conference config · Places2 256x256 (mask 10-60%) | FID: 75.625 -> 22.121 | Improved |
| add random quantization during training (ablation) | PUT conference config · Places2 256x256 (mask 10-60%) | FID: 44.588 -> 22.121 | Improved |
| increase codebook K from 512 to 8192 (ablation) | PUT · Places2 256x256 (mask 10-60%) | FID: 22.696 -> 23.921 | Hurt |
| add Gumbel-softmax relaxation and mask embedding (ablation) | PUT K=8192 · Places2 256x256 (mask 10-60%) | FID: 23.921 -> 21.538 | Improved |
| add mask embedding (ablation) | PUT K=8192 with Gumbel-softmax · Places2 256x256 (mask 10-60%) | FID: 24.179 -> 21.538 | Improved |
| train with conditions, give semantic and structural maps (ablation) | PUT · ImageNet 256x256 (mask 10-60%) | FID: 26.402 -> 17.916 | Improved |
| train with conditions but give none at inference (ablation) | PUT · Places2 256x256 (mask 10-60%) | FID: 21.234 -> 21.851 | No clear effect |
| replace 20 removed categories with unknown categories (ablation) | PUT Sem.113 · ImageNet 256x256 (mask 10-60%) | FID: 24.877 -> 23.674 | Improved |
Closest papers by meaning
- Reduce Information Loss in Transformers for Pluralistic Image Inpainting (CVPR 2022 · similarity 0.95)
- Don't Look into the Dark: Latent Codes for Pluralistic Image Inpainting (CVPR 2024 · similarity 0.86)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.85)
- Outlier-aware Slicing for Post-Training Quantization in Vision Transformer (ICML 2024 · similarity 0.84)
- Image and Video Tokenization with Binary Spherical Quantization (ICLR 2025 · similarity 0.84)
- SyFormer: Structure-Guided Synergism Transformer for Large-Portion Image Inpainting (AAAI 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.
