Research map / Image generation / Image restoration
Localization of Deep Inpainting Using High-Pass Fully Convolutional Network
Haodong Li, Jiwu Huang
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 |
|---|---|---|---|
| use transpose convolution instead of bilinear resizing for up-sampling (ablation) | high-pass FCN · Synthetic inpainted ImageNet (QF=75, ablation) | F1-score (%): 89.13 -> 91.53 | Improved |
| use focal loss instead of standard cross entropy (ablation) | high-pass FCN · Synthetic inpainted ImageNet (QF=75, ablation) | F1-score (%): 91.17 -> 91.53 | Improved |
| use focal loss instead of weighted cross entropy (ablation) | high-pass FCN · Synthetic inpainted ImageNet (QF=75, ablation) | F1-score (%): 90.69 -> 91.53 | Improved |
| use weighted cross entropy instead of focal loss (ablation) | high-pass FCN · Synthetic inpainted ImageNet (QF=75, ablation) | Recall (%): 89.84 -> 90.54 | Mixed |
| initialize pre-filter with 1st-order derivative high-pass filters rather than higher-order ones (ablation) | high-pass FCN · Synthetic inpainted ImageNet (QF=75, ablation) | F1-score (%) | Improved |
| add high-pass pre-filtering module (ablation) | high-pass FCN · Synthetic inpainted ImageNet (QF=75, ablation) | F1-score (%) | Improved |
| add Gaussian noise at SNR 40 dB instead of 50 dB | Proposed · Noisy JPEG images, QF=75 | F1-score (%): 91.49 -> 91.18 | No clear effect |
| add Gaussian noise at SNR 30 dB instead of 50 dB | Proposed · Noisy JPEG images, QF=75 | F1-score (%): 91.49 -> 77.00 | Hurt |
| lower JPEG quality factor from 96 to 75 | Proposed · Synthetic inpainted ImageNet | F1-score (%): 97.28 -> 91.53 | Hurt |
| Proposed instead of MFCN splicing localizer | MFCN · Synthetic inpainted ImageNet (QF=75) | F1-score (%): 45.01 -> 91.53 | Improved |
| Proposed instead of best Patch-CNN (b=64) | Patch-CNN · Synthetic inpainted ImageNet (QF=75) | F1-score (%): 75.15 -> 91.53 | Improved |
| increase Patch-CNN window size from 16 to 32 then 64 | Patch-CNN · Synthetic inpainted ImageNet (QF=96) | F1-score (%): 78.49 -> 82.90 -> 72.80 | Mixed |
Closest papers by meaning
- Deep Learning-Based Image and Video Inpainting: A Survey (IJCV 2024 · similarity 0.89)
- High-Resolution Image Inpainting with Iterative Confidence Feedback and Guided Upsampling (ECCV 2020 · similarity 0.89)
- Image Inpainting With Learnable Bidirectional Attention Maps (ICCV 2019 · similarity 0.89)
- Learning Prior Feature and Attention Enhanced Image Inpainting (ECCV 2022 · similarity 0.88)
- Internal Video Inpainting by Implicit Long-Range Propagation (ICCV 2021 · similarity 0.87)
- InpDiffusion: Image Inpainting Localization via Conditional Diffusion Models (AAAI 2025 · similarity 0.87)
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.
