Research map / Detection and segmentation / CNN features
Context Encoders: Feature Learning by Inpainting
Check 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 |
|---|---|---|---|
| Pretrain AlexNet by predicting missing image context | AlexNet classifier · PASCAL VOC 2007 classification | mAP: 53.8% -> 56.5% | Helped |
| Pretrain AlexNet by predicting missing image context | Fast R-CNN detector · PASCAL VOC 2007 object detection | mAP: 41.9% -> 44.5% | Helped |
| Pretrain AlexNet by predicting missing image context | FCN · PASCAL VOC 2012 semantic segmentation | score: 25.2% -> 30.0% | Helped |
| Use context encoder features instead of HOG for nearest-neighbor inpainting | NN-inpainting · Paris StreetView held-out semantic inpainting | Mean L1 Loss: 19.92% -> 15.10% | Helped |
| Use joint reconstruction and adversarial loss instead of HOG-feature nearest-neighbor inpainting | Semantic inpainting · Paris StreetView held-out | PSNR: 12.79 dB -> 18.58 dB | Helped |
| Add adversarial loss to reconstruction loss (ablation) | Context Encoder · Semantic inpainting | visual sharpness | Helped |
| Use reconstruction loss without adversarial loss (ablation) | Context Encoder · Semantic inpainting | visual sharpness | Hurt |
| Use adversarial loss without reconstruction loss (ablation) | Context Encoder · Semantic inpainting | context coherence | Hurt |
| Use joint reconstruction and adversarial loss instead of either loss alone (ablation) | Context Encoder · Semantic inpainting | inpainting quality | Helped |
| Replace L2 reconstruction loss with L1 reconstruction loss (ablation) | Context Encoder · Image reconstruction | reconstruction quality | No clear effect |
| Use random block or random region masks instead of a fixed central mask (ablation) | Context Encoder feature learning · Transferable visual features | feature generality | Helped |
| Replace pooling layers with convolutions (ablation) | Context Encoder · Image inpainting | inpainting detail | Helped |
Closest papers by meaning
- Learning Pyramid-Context Encoder Network for High-Quality Image Inpainting (CVPR 2019 · similarity 0.90)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.90)
- Generative Image Inpainting With Contextual Attention (CVPR 2018 · similarity 0.89)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.86)
- A Simple Image Segmentation Framework via In-Context Examples (NeurIPS 2024 · similarity 0.86)
- Explore In-Context Segmentation via Latent Diffusion Models (AAAI 2025 · similarity 0.86)
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.