Research map / Video / Video segmentation and tracking
ReferEverything: Towards Segmenting Everything We Can Speak of in Videos
Anurag Bagchi, Zhipeng Bao, Yu-Xiong Wang, Pavel Tokmakov, Martial Hebert
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 |
|---|---|---|---|
| supervise mask prediction in the VAE latent space instead of RGB pixels (256px) (ablation) | REM (MS-1.4B) · Ref-VPS | J: 31.6 -> 35.8 | Improved |
| supervise in latent space instead of RGB with fine-tuned VAE (256px) (ablation) | REM (MS-1.4B) · Ref-YTB | J&F: 60.4 -> 61.4 | Improved |
| keep the pre-trained frozen VAE decoder instead of a CNN mask decoder (ablation) | REM (MS-1.4B) · Ref-VPS | J: 29.4 -> 40.0 | Improved |
| keep the pre-trained frozen VAE decoder instead of an MLP mask decoder (ablation) | REM (MS-1.4B) · Ref-VPS | J: 33.1 -> 40.0 | Improved |
| add image-only first stage (two-stage training) (ablation) | REM (MS-1.4B) · Ref-VPS | J: 46.33 -> 49.0 | Improved |
| add image data in second stage (ablation) | REM (MS-1.4B) · Ref-VPS | J: 40.7 -> 49.0 | Improved |
| add image data in second stage (ablation) | REM (MS-1.4B) · Ref-YTB | J&F: 68.7 -> 68.4 | No clear effect |
| double the image data (2:1 images to videos) (ablation) | REM (MS-1.4B) · Ref-VPS | J: 49.0 -> 48.0 | Hurt |
| use noise timestep 0 instead of 50 (ablation) | REM · Ref-YTB | J: 35.0 -> 40.5 | Improved |
| use noise timestep 0 instead of 200 (ablation) | REM · Ref-YTB | J&F: 59.2 -> 62.9 | Improved |
| use ModelScope backbone instead of Stable Diffusion 2.1 image model (ablation) | REM · Ref-VPS | J: 29.8 -> 40.0 | Improved |
| use VideoCrafter-2 with curated data instead of VideoCrafter-1 (ablation) | REM · Ref-VPS | J: 28.0 -> 36.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- GLUS — GLUS: Global-Local Reasoning Unified into A Single Large Language Model for Video Segmentation (CVPR 2025) · better on 2 of 2 shared results
- VISA-13B — VISA: Reasoning Video Object Segmentation via Large Language Model (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exploring Pre-trained Text-to-Video Diffusion Models for Referring Video Object Segmentation (ECCV 2024 · similarity 0.92)
- Deforming Videos to Masks: Flow Matching for Referring Video Segmentation (ICLR 2026 · similarity 0.92)
- Temporal Collection and Distribution for Referring Video Object Segmentation (ICCV 2023 · similarity 0.91)
- LoSh: Long-Short Text Joint Prediction Network for Referring Video Object Segmentation (CVPR 2024 · similarity 0.90)
- SAMWISE: Infusing Wisdom in SAM2 for Text-Driven Video Segmentation (CVPR 2025 · similarity 0.89)
- SetCon: Towards Open-Ended Referring Segmentation via Set-Level Concept Prediction (NeurIPS 2026 · similarity 0.89)
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.
