Research map / Detection and segmentation / Open-set and few-shot detection
Semantic Projection Network for Zero- and Few-Label Semantic Segmentation
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 |
|---|---|---|---|
| Replace hinge visual-semantic loss with cross-entropy | SPNet on COCO-Stuff · COCO-Stuff ZLSS | mean IoU (unseen classes): 31.8 -> 35.2 | Helped |
| Combine fastText and word2vec embeddings (ablation) | SPNet on COCO-Stuff · COCO-Stuff ZLSS | mean IoU (unseen classes): 33.1 and 32.1 -> 35.2 | Helped |
| Use fastText rather than combined fastText and word2vec (ablation) | SPNet on COCO-Stuff · COCO-Stuff ZLSS | mean IoU (unseen classes): 35.2 -> 33.1 | Hurt |
| Use word2vec rather than combined fastText and word2vec (ablation) | SPNet on COCO-Stuff · COCO-Stuff ZLSS | mean IoU (unseen classes): 35.2 -> 32.1 | Hurt |
| Replace VGG16 with ResNet101 backbone (ablation) | SPNet · COCO-Stuff ZLSS | mean IoU (unseen classes): 26.3 -> 35.2 | Helped |
| Replace VGG16 with ResNet101 backbone (ablation) | SPNet · PASCAL-VOC ZLSS | mean IoU (unseen classes): 47.4 -> 49.5 | Helped |
| Calibrate scores by reducing seen-class predictions (ablation) | SPNet on COCO-Stuff · COCO-Stuff GZLSS | mean IoU (unseen classes): 0.2% -> 8.33% | Helped |
| Calibrate scores by reducing seen-class predictions (ablation) | SPNet on PASCAL-VOC · PASCAL-VOC GZLSS | mean IoU (unseen classes): 0.01% -> 29.33% | Helped |
| Use SPNet instead of fine-tuned DeepLab baseline | COCO-Stuff with 1 labeled sample per novel class · COCO-Stuff FLSS | mean IoU (novel classes): 27.69% -> 47.90% | Helped |
| Use SPNet instead of fine-tuned DeepLab baseline | PASCAL-VOC with 1 labeled sample per novel class · PASCAL-VOC FLSS | mean IoU (novel classes): 29.17% -> 71.52% | Helped |
| Increase labeled samples per novel class from 1 to 5 (ablation) | SPNet on COCO-Stuff and PASCAL-VOC · FLSS | mean IoU (novel classes): ≈20% mIoU | Helped |
| Use SPNet instead of DeepLab baseline | Few-label segmentation across COCO-Stuff and PASCAL-VOC · GFLSS | overall mean IoU: 21.24% mIoU | Helped |
Closest papers by meaning
- Single-Stage Semantic Segmentation From Image Labels (CVPR 2020 · similarity 0.91)
- Zero-Shot Semantic Segmentation (NeurIPS 2019 · similarity 0.91)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.89)
- Active Learning for Semantic Segmentation with Multi-class Label Query (NeurIPS 2023 · similarity 0.89)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.89)
- A Surprisingly Simple Approach to Generalized Few-Shot Semantic Segmentation (NeurIPS 2024 · 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.