Research map / Image generation / GANs
InSeGAN: A Generative Approach to Segmenting Identical Instances in Depth Images
Anoop Cherian, Gonçalo Dias Pais, Siddarth Jain, Tim K. Marks, Alan Sullivan
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 |
|---|---|---|---|
| add the 3D generator modules (implicit template, pose decoder, STN) instead of 2D convolutions (ablation) | InSeGAN · Insta-10 (avg) | mIoU: 0.419 -> 0.590 | Improved |
| use optimal transport instead of greedy alignment (ablation) | InSeGAN · Insta-10 Bolt | mIoU: 0.383 -> 0.424 | Improved |
| use optimal transport instead of greedy alignment (ablation) | InSeGAN · Insta-10 Obj01 | mIoU: 0.664 -> 0.686 | Improved |
| add the re-generation loss L_p (ablation) | InSeGAN · Insta-10 Bolt | mIoU: 0.312 -> 0.424 | Improved |
| add the re-generation loss L_p (ablation) | InSeGAN · Insta-10 Obj01 | mIoU: 0.360 -> 0.686 | Improved |
| add the intermediate feature loss L_i (ablation) | InSeGAN · Insta-10 Bolt | mIoU: 0.303 -> 0.312 | No clear effect |
| add the intermediate feature loss L_i (ablation) | InSeGAN · Insta-10 Obj01 | mIoU: 0.402 -> 0.360 | Hurt |
| add Gaussian noise sigma 0.5 to input depth images (ablation) | InSeGAN · Insta-10 Bolt | mIoU: 0.424 -> 0.318 | Hurt |
| add Gaussian noise sigma 0.5 to input depth images (ablation) | InSeGAN · Insta-10 Obj01 | mIoU: 0.686 -> 0.421 | Hurt |
| train on more samples (500 up to 9800) (ablation) | InSeGAN · Insta-10 | mIoU | Improved |
| set number of instances n far from the true 5 (ablation) | InSeGAN · Insta-10 Bolt and Obj01 | mIoU | Hurt |
Closest papers by meaning
- PrimitiveNet: Primitive Instance Segmentation With Local Primitive Embedding Under Adversarial Metric (ICCV 2021 · similarity 0.87)
- InstaGAN: Instance-aware Image-to-Image Translation (ICLR 2019 · similarity 0.87)
- Learning Segmentation Masks with the Independence Prior (AAAI 2019 · similarity 0.86)
- 3D generation on ImageNet (ICLR 2023 · similarity 0.86)
- UniC-Lift: Unified 3D Instance Segmentation via Contrastive Learning (AAAI 2026 · similarity 0.86)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.85)
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.
