Research map / Video / Video segmentation and tracking
DC-SAM: In-Context Segment Anything in Images and Videos via Dual Consistency
Mengshi Qi, Pengfei Zhu, Xiangtai Li, Xiaoyang Bi, Lu Qi, Huadóng Ma, Ming–Hsuan Yang
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 positive and negative prompt branches (ablation) | VRP-SAM baseline · PASCAL-5i (ResNet50) | mIoU: 71.9 -> 72.1 (+0.2) | Improved |
| add SAM feature fusion (ablation) | VRP-SAM + Pos-Neg branch · PASCAL-5i (ResNet50) | mIoU: 72.1 -> 72.8 (+0.6) | Improved |
| add cyclic consistency (ablation) | DC-SAM · PASCAL-5i (ResNet50) | mIoU: 72.8 -> 73.0 (+0.2) | Improved |
| pre-train the prompt generator on images before video fine-tuning (ablation) | DC-SAM · IC-VOS | J&F: 61.54 -> 67.23 | Improved |
| also train SAM2 decoder during image pre-training (ablation) | DC-SAM · IC-VOS | J&F: 67.23 -> 70.59 | Improved |
| fine-tune the memory module instead of only the decoder (ablation) | DC-SAM · IC-VOS | J&F: 70.59 -> 71.23 | Improved |
| fine-tune both decoder and memory module (ablation) | DC-SAM · IC-VOS | J&F: 71.23 -> 71.52 | Improved |
| fine-tune with mask tubes instead of images (ablation) | DC-SAM · IC-VOS | J&F: 66.94 -> 70.59 | Improved |
| use LoRA instead of full fine-tuning of SAM2 parts (ablation) | DC-SAM · IC-VOS | J&F: 71.52 -> 69.79; 11.26M -> 0.11M params | Mixed |
| increase queries per branch up to 25 (ablation) | DC-SAM · PASCAL-5i | mIoU | Mixed |
| move from 1-shot to 5-shot support | DC-SAM · PASCAL-5i (ResNet-50) | mIoU: 73.0 -> 76.5 | Improved |
| move from 1-shot to 5-shot support | DC-SAM · PASCAL-5i (VGG-16) | mIoU: 70.4 -> 74.3 | Improved |
Closest papers by meaning
- SAM-I2V: Upgrading SAM to Support Promptable Video Segmentation with Less than 0.2% Training Cost (CVPR 2025 · similarity 0.90)
- Stable Segment Anything Model (ICLR 2025 · similarity 0.90)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · similarity 0.89)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.89)
- Personalize Segment Anything Model with One Shot (ICLR 2024 · similarity 0.89)
- VRP-SAM: SAM with Visual Reference Prompt (CVPR 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.
