Research map / Detection and segmentation / Weakly supervised detection and segmentation
Sparse Object-Level Supervision for Instance Segmentation With Pixel Embeddings
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 |
|---|---|---|---|
| Apply a differentiable object-level loss to pixel embeddings (ablation) | SPOCO · CVPPP validation | SBD: 0.847 -> 0.852 | Helped |
| Add embedding consistency to the object-level loss (ablation) | SPOCO · Cityscapes validation | mAP@0.5: 0.418 -> 0.459 | Helped |
| Add embedding consistency to the object-level loss (ablation) | SPOCO · CVPPP validation | SBD: 0.852 -> 0.849 | Hurt |
| Add unlabeled push loss (ablation) | SPOCO@0.1 · CVPPP A1 test | SBD: 0.734 -> 0.788 | Helped |
| Add embedding consistency loss (ablation) | SPOCO@0.1 · CVPPP A1 test | SBD: 0.720 -> 0.788 | Helped |
| Add embedding consistency loss (ablation) | SPOCO@0.4 · CVPPP A1 test | SBD: 0.738 -> 0.824 | Helped |
| Add unlabeled push loss (ablation) | SPOCO@0.8 · CVPPP A1 test | SBD: 0.797 -> 0.828 | Helped |
| Use an exponential-moving-average consistency network instead of shared, dropout, or independently trained weights (ablation) | SPOCO@0.1 · CVPPP validation | SBD: Shared 0.602 $\pm$ 0.016; Dropout 0.507 $\pm$ 0.060; Trained 0.591 $\pm$ 0.131; Momentum 0.649 $\pm$ 0.045 | Helped |
| Add embedding consistency during target-domain fine-tuning (ablation) | VNC-MitoEM SPOCO@0.05 · MitoEM test | mAP: 0.227 $\pm$ 0.002 -> 0.277 $\pm$ 0.006 | Helped |
| Add adversarial WGAN loss to the object-level objective (ablation) | SPOCO · Arabidopsis Ovules test | Adapted Rand error: 0.048 -> 0.042 | Helped |
| Train with only 10% of object masks rather than full supervision | SPOCO · Arabidopsis Ovules test | Adapted Rand error: 0.048 -> 0.069 | Hurt |
| Fine-tune the source model using 5% of target object masks | Ovules-trained U-Net · Stem test | Adapted Rand error: 0.227 -> 0.109 $\pm$ 0.002 | Helped |
Closest papers by meaning
- Learning to Model Pixel-Embedded Affinity for Homogeneous Instance Segmentation (AAAI 2022 · similarity 0.88)
- Object-Guided Instance Segmentation for Biological Images (AAAI 2020 · similarity 0.88)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.87)
- Instance Segmentation of Biological Images Using Harmonic Embeddings (CVPR 2020 · similarity 0.87)
- Unified Mask Embedding and Correspondence Learning for Self-Supervised Video Segmentation (CVPR 2023 · similarity 0.87)
- Training-Free Dataset Pruning for Instance Segmentation (ICLR 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.