Research map / Image generation / Autoregressive and latent generation
A Layer-Based Sequential Framework for Scene Generation with GANs
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 sequential background-then-foreground composition instead of non-sequential conditional generation | Scene generation on MS-COCO · MS-COCO train subset | Mean IoU: 0.608 -> 0.650 | Helped |
| Use sequential background-then-foreground composition instead of non-sequential conditional generation | Scene generation on MS-COCO · MS-COCO validation subset | Mean IoU: 0.605 -> 0.650 | Helped |
| Replace the generated background with a Pix2Pix++ background (ablation) | Sequential · MS-COCO train subset | FID: 28.7 -> 23.2 | Helped |
| Add adversarial training to the L1 reconstruction baseline | Standard CNN · MS-COCO train subset | Mean IoU: 0.298 -> 0.504 | Helped |
| Add adversarial training to the L1 reconstruction baseline | Standard CNN · MS-COCO validation subset | Mean IoU: 0.272 -> 0.480 | Helped |
| Add enhanced GAN training techniques, including spectral normalization and feature matching | Pix2Pix · MS-COCO train subset | Mean IoU: 0.504 -> 0.608 | Helped |
| Add enhanced GAN training techniques, including spectral normalization and feature matching | Pix2Pix · MS-COCO validation subset | Mean IoU: 0.480 -> 0.605 | Helped |
| Use the proposed sequential model without replacing its background generator | Scene generation on MS-COCO · MS-COCO train subset | FID: 24.0 -> 28.7 | Hurt |
| Generate foreground objects sequentially using object masks | Affine transformations of foreground masks · MS-COCO scene generation | Preservation of background and foreground under transformation | Helped |
| Generate foreground objects sequentially using object masks | Foreground object occlusion · MS-COCO scene generation | Occlusion artifact robustness | Helped |
Closest papers by meaning
- LR-GAN: Layered Recursive Generative Adversarial Networks for Image Generation (ICLR 2017 · similarity 0.89)
- Compositional Transformers for Scene Generation (NeurIPS 2021 · similarity 0.89)
- Exploiting Relationship for Complex-scene Image Generation (AAAI 2021 · similarity 0.89)
- Local Class-Specific and Global Image-Level Generative Adversarial Networks for Semantic-Guided Scene Generation (CVPR 2020 · similarity 0.88)
- Generating Multiple Objects at Spatially Distinct Locations (ICLR 2019 · similarity 0.88)
- Object-Centric Image Generation from Layouts (AAAI 2021 · similarity 0.88)
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.