Research map / Detection and segmentation / Open-set and few-shot detection
TopNet: Transformer-Based Object Placement Network for Image Compositing
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 dense transformer-based local object-background correlation instead of global-feature regression | TopNet · Pixabay | Mean IOU (top-5): 0.448 -> 0.620 | Helped |
| use dense transformer-based local object-background correlation instead of PlaceNet | TopNet · OPA | Mean IOU (top-5): 0.225 -> 0.241 | Helped |
| use TopNet instead of the sliding-window retrieval model | Pixabay · Pixabay | inference time: 1.69 -> 0.11 | Helped |
| use TopNet for location prediction with ground-truth scale instead of the classifier | TopNet · Pixabay, ground-truth scale | Mean NS: 0.70 -> 0.90 | Helped |
| use TopNet for scale prediction with ground-truth location instead of retrieval | TopNet · Pixabay, ground-truth location | Mean Error: 0.105 -> 0.052 | Helped |
| use TopNet rather than the strongest baseline in the real-world user study | TopNet · Real-world images | Satisfactory: 35.8 -> 39.4 | Helped |
| add sparse contrastive loss instead of binary assignment (ablation) | TopNet · Pixabay | Mean IOU (top-5): 0.050 -> 0.620 | Helped |
| add sparse contrastive loss instead of Gaussian assignment (ablation) | TopNet · Pixabay | Mean IOU (top-5): 0.252 -> 0.620 | Helped |
| add transformer attention between local background and global object features rather than use only global features (ablation) | TopNet · Pixabay | Mean IOU (top-5): 0.069 -> 0.620 | Helped |
| add transformer attention rather than directly concatenate the object feature at each background location (ablation) | TopNet · Pixabay | Mean IOU (top-5): 0.466 -> 0.620 | Helped |
Closest papers by meaning
- BOOTPLACE: Bootstrapped Object Placement with Detection Transformers (CVPR 2025 · similarity 0.88)
- MonoPlace3D: Learning 3D-Aware Object Placement for 3D Monocular Detection (CVPR 2025 · similarity 0.84)
- $\textit{Bifr\"ost}$: 3D-Aware Image Compositing with Language Instructions (NeurIPS 2024 · similarity 0.84)
- Add-it: Training-Free Object Insertion in Images With Pretrained Diffusion Models (ICLR 2025 · similarity 0.84)
- DOCTR: Disentangled Object-Centric Transformer for Point Scene Understanding (AAAI 2024 · similarity 0.83)
- MOVIS: Enhancing Multi-Object Novel View Synthesis for Indoor Scenes (CVPR 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.