Research map / Video / Video segmentation and tracking
TubeFormer-DeepLab: Video Mask Transformer
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 |
|---|---|---|---|
| Use direct prediction of class-labeled video tubes in a unified mask-transformer model | Video segmentation · KITTI-STEP test | STQ: +13.1 STQ | Helped |
| Use direct class-labeled tube prediction | Video segmentation · VSPW test | mIoU: +21 mIoU | Helped |
| Use TubeFormer-DeepLab rather than IFC | YouTube-VIS-2019 · YouTube-VIS-2019 val | AP: +2.9 AP | Helped |
| Add a depth prediction branch | TubeFormer-DeepLab · SemKITTI-DVPS test | DSTQ: +3.4 DSTQ | Helped |
| Add frame-latent attention using latent memory (ablation) | TubeFormer-DeepLab-Simple · KITTI-STEP val | STQ: +1.7 STQ | Helped |
| Add memory-to-latent attention (ablation) | TubeFormer-DeepLab · KITTI-STEP val | STQ | No clear effect |
| Add latent-to-memory attention (ablation) | TubeFormer-DeepLab · KITTI-STEP val | STQ | No clear effect |
| Set latent memory size to 8 rather than 16 (ablation) | TubeFormer-DeepLab · KITTI-STEP val | STQ: 70.03 -> 69.39 | Hurt |
| Set latent memory size to 32 rather than 16 (ablation) | TubeFormer-DeepLab · KITTI-STEP val | STQ: 70.03 -> 69.57 | Hurt |
| Add temporal consistency loss between overlapping clips (ablation) | TubeFormer-DeepLab · KITTI-STEP val | STQ: +0.5 STQ | Helped |
| Add clip-level copy-paste augmentation (ablation) | TubeFormer-DeepLab · KITTI-STEP val | STQ: +0.9 STQ | Helped |
| Add ImageNet-22K pretraining (ablation) | TubeFormer-DeepLab B1 · KITTI-STEP val | STQ: +1.6 STQ | Helped |
Closest papers by meaning
- MED-VT: Multiscale Encoder-Decoder Video Transformer With Application To Object Segmentation (CVPR 2023 · similarity 0.86)
- Unified Mask Embedding and Correspondence Learning for Self-Supervised Video Segmentation (CVPR 2023 · similarity 0.85)
- Mask Propagation for Efficient Video Semantic Segmentation (NeurIPS 2023 · similarity 0.85)
- DTMFormer: Dynamic Token Merging for Boosting Transformer-Based Medical Image Segmentation (AAAI 2024 · similarity 0.85)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.84)
- PEM: Prototype-based Efficient MaskFormer for Image Segmentation (CVPR 2024 · similarity 0.84)
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.