Research map / Video / Video segmentation and tracking
Video K-Net: A Simple, Strong, and Unified Baseline for Video 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 |
|---|---|---|---|
| Add kernel association embedding (KAE) (ablation) | K-Net · KITTI-STEP validation | STQ: 67.5 -> 69.3 | Helped |
| Add kernel linking (KL) after KAE (ablation) | K-Net with KAE · KITTI-STEP validation | STQ: 69.3 -> 70.2 | Helped |
| Add kernel fusion (KF) after KAE and KL (ablation) | K-Net with KAE and KL · KITTI-STEP validation | STQ: 70.2 -> 70.9 | Helped |
| Add external Mask-Emb appearance features to kernel embeddings (ablation) | Video K-Net · KITTI-STEP validation | STQ: 70.9 -> 70.8 | No clear effect |
| Use proposal sampling from prior work instead of ground-truth-matched kernels (ablation) | K-Net association training · KITTI-STEP validation | STQ: 69.3 -> 63.1 | Hurt |
| Move linking and fusion from stage 3 to stage 2 (ablation) | Video K-Net · KITTI-STEP validation | STQ: 70.9 -> 68.5 | Hurt |
| Move linking and fusion from stage 3 to stage 1 (ablation) | Video K-Net · KITTI-STEP validation | STQ: 70.9 -> 66.9 | Hurt |
| Train on the key frame only instead of jointly training both frames (ablation) | Video K-Net · KITTI-STEP validation | STQ: 70.9 -> 70.1 | Hurt |
| Remove kernel updating before kernel fusion (ablation) | Video K-Net · KITTI-STEP validation | STQ: 70.9 -> 67.1 | Hurt |
| Use Swin-base instead of ResNet50 | Video K-Net · KITTI-STEP validation | STQ: 0.71 -> 0.73 | Helped |
| Use Video K-Net instead of VPSNet | Cityscapes-VPS · Cityscapes-VPS validation | VPQ Average: 57.5 -> 57.8 | Helped |
| Add Video K-Net kernel-based video processing to DeepLabv3+ | DeepLabv3+ · VSPW validation | mIoU: 35.7 -> 37.9 | Helped |
Closest papers by meaning
- K-Net: Towards Unified Image Segmentation (NeurIPS 2021 · similarity 0.93)
- STEP: Segmenting and Tracking Every Pixel (NeurIPS 2021 · similarity 0.89)
- Video Panoptic Segmentation (CVPR 2020 · similarity 0.89)
- NewsNet: A Novel Dataset for Hierarchical Temporal Segmentation (CVPR 2023 · similarity 0.89)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.88)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 2025 · similarity 0.87)
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.