Research map / Video / Video segmentation and tracking
CLEAR: Context-Aware Learning with End-to-End Mask-Free Inference for Adaptive Video Subtitle Removal
Qingdong He, Chaoyi Wang, Peng Tang, Yifan Yang, Xiaobin Hu
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 Stage I prior with focal weighting to LoRA-only baseline (ablation) | LoRA-only baseline · Chinese subtitle test set (400 samples) | PSNR: 21.62 -> 23.11 | Improved |
| add Stage I prior with focal weighting to LoRA-only baseline (ablation) | LoRA-only baseline · Chinese subtitle test set (400 samples) | VFID: 34.74 -> 38.21 | Hurt |
| add context distillation (M2) (ablation) | CLEAR · Chinese subtitle test set (400 samples) | PSNR: 23.11 -> 24.72 | Improved |
| add context-aware adaptation with generation feedback (M3) (ablation) | CLEAR · Chinese subtitle test set (400 samples) | PSNR: 24.72 -> 25.09 | Improved |
| add context consistency (L1 sparsity + KL) loss (M4) (ablation) | CLEAR · Chinese subtitle test set (400 samples) | PSNR: 25.09 -> 26.80 | Improved |
| add context consistency loss (M4) (ablation) | CLEAR · Chinese subtitle test set (400 samples) | VFID: 31.56 -> 20.37 | Improved |
| use 10 denoising steps instead of 5 (ablation) | CLEAR · Chinese subtitle test set (400 samples) | PSNR / TWE: PSNR 26.80 -> 29.43; TWE 1.227 -> 1.358 | Mixed |
| raise CFG scale from 1.0 to 1.2 (ablation) | CLEAR · Chinese subtitle test set (400 samples) | VFID: 20.37 -> 40.71 | Hurt |
| lower LoRA scale from 1.0 to 0.5 (ablation) | CLEAR · Chinese subtitle test set (400 samples) | LPIPS: 0.101 -> 0.184 | Hurt |
| raise LoRA scale from 1.0 to 1.5 (ablation) | CLEAR · Chinese subtitle test set (400 samples) | VFID: 20.37 -> 42.16 | Hurt |
| replace mask-guided baselines with mask-free CLEAR | CLEAR vs MiniMax-Remover · Chinese subtitle test set (400 samples) | PSNR: 20.03 -> 26.80 | Improved |
| replace mask-guided baselines with mask-free CLEAR | CLEAR vs ProPainter · Chinese subtitle test set (400 samples) | Flow Var: 0.885 -> 0.029 | 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.
- MiniMax-Remover — MiniMax-Remover: Taming Bad Noise Helps Video Object Removal (NeurIPS 2025) · better on 6 of 6 shared results
- ProPainter — ProPainter: Improving Propagation and Transformer for Video Inpainting (ICCV 2023) · better on 5 of 6 shared results
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
- Cluster-Wise Spatio-Temporal Masking for Efficient Video-Language Pretraining (CVPR 2026 · similarity 0.85)
- Learning Stochastic Bridges for Video Object Removal via Video-to-Video Translation (ICML 2026 · similarity 0.85)
- Masked Diffusion Vision-Language Models for Temporal Action Localization (NeurIPS 2026 · similarity 0.84)
- LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025 · similarity 0.84)
- DSAQuant: Denoising-Stage-Aligned Quantization-Aware Training for Video Generation (NeurIPS 2026 · similarity 0.84)
- A Simple Recipe for Contrastively Pre-training Video-First Encoders Beyond 16 Frames (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.
