Research map / Image generation / GANs
The Missing Data Encoder: Cross-Channel Image Completion with Hide-and-Seek Adversarial Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use MDE-REC for face completion instead of GFC | CelebA face completion · right half, both eyes, and eye occlusions | pSNR: 18.3/19.1/18.9 -> 21.8/22.4/22.6 | Helped |
| use MDE-REC for mouth completion instead of GFC | CelebA face completion · mouth | SSIM: 0.824 -> 0.818 | Hurt |
| train with independent random masks for each color channel (REC) rather than RE (ablation) | MDE representation learning · CelebA landmark localization | average point-to-point error: 2.039 -> 1.509 | Helped |
| train with independent random masks for each color channel (REC) rather than RE (ablation) | MDE representation learning · CelebA attribute prediction | average trace of confusion matrices: 12.72 -> 10.49 | Helped |
| use channel-independent REC masks rather than fixed reverse-inpainting masks (ablation) | MDE · CelebA transfer tasks | landmark localization and attribute prediction: 1,496 -> 1.451; 13,73 -> 10.16 | Mixed |
| use adversarial and perceptual losses instead of reconstruction loss alone (ablation) | MDE-REC (S=0.1) · CelebA image completion | inception score | Helped |
| add hide-and-seek coordinate loss to perceptual and adversarial losses (ablation) | MDE-REC (S=0.1) · CelebA image completion | inception score | Helped |
| train with reconstruction loss only, without adversarial/perceptual objectives (ablation) | MDE-REC · CelebA image completion | inception score | Hurt |
| use lower masking ratio S=0.1 rather than higher S (ablation) | MDE-RE and MDE-REC · CelebA completion | inception score | Helped |
| train with REC rather than RE (ablation) | MDE · colorization and REC test tasks | pSNR and inception score | Helped |
| train with a task-specific mask rather than REC (ablation) | MDE · mismatched completion scenario | pSNR and inception score | Hurt |
Closest papers by meaning
- Large Scale Image Completion via Co-Modulated Generative Adversarial Networks (ICLR 2021 · similarity 0.88)
- Pluralistic Image Completion (CVPR 2019 · similarity 0.87)
- Generative Face Completion (CVPR 2017 · similarity 0.87)
- Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion (CVPR 2025 · similarity 0.84)
- MaGIC: Multi-modality Guided Image Completion (ICLR 2024 · similarity 0.83)
- Efficient Indoor Depth Completion Network Using Mask-adaptive Gated Convolution (AAAI 2025 · similarity 0.83)
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.