Research map / Video / Video-language understanding
Visual Text Correction
Amir Mazaheri, Mubarak Shah
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 word position gating to convolutional n-grams (ablation) | Conv N-Grams · LSMDC-based VTC detection | detection accuracy (%): 66.8 -> 69.0 | Improved |
| add bidirectional (BiLSTM) instead of one-way LSTM (ablation) | long-term dependencies · LSMDC-based VTC detection | detection accuracy (%): 58.0 -> 67.2 | Improved |
| combine conv n-grams with long-term dependencies (ablation) | text-only detector · LSMDC-based VTC detection | detection accuracy (%): 69.0 -> 72.5 | Improved |
| add visual gated bias (ablation) | text-only detector · LSMDC-based VTC detection | detection accuracy (%): 72.5 -> 74.5 | Improved |
| use visual gated bias instead of feature concatenation (ablation) | detector with video · LSMDC-based VTC detection | detection accuracy (%): 72.8 -> 74.5 | Improved |
| replace vanilla LSTM with proposed text model | detector · LSMDC-based VTC detection | detection accuracy (%): 28.0 -> 72.5 | Improved |
| add C3D video features (ablation) | text-only corrector · LSMDC-based VTC correction | correction accuracy (%): 35.2 -> 38.6 | Improved |
| add VGG19 features instead of C3D (ablation) | corrector with video · LSMDC-based VTC correction | correction accuracy (%): 38.6 -> 38.8 | Improved |
| add both VGG19 and C3D (ablation) | corrector with video · LSMDC-based VTC correction | correction accuracy (%): 38.8 -> 38.9 | No clear effect |
| train end-to-end instead of detector plus pre-trained FIB (ablation) | corrector · LSMDC-based VTC correction | correction accuracy (%): 36.0 -> 38.9 | Improved |
| add video to text-only model with multiple inaccuracies (ablation) | detector · Multiple inaccuracies, all k (detection) | sentence-based accuracy (%): 59 -> 63 | Improved |
| add video to text-only model with multiple inaccuracies (ablation) | corrector · Multiple inaccuracies, all k (correction) | word-based accuracy (%): 3 -> 18 | Improved |
Closest papers by meaning
- VCR: A Task for Pixel-Level Complex Reasoning in Vision Language Models via Restoring Occluded Text (ICLR 2025 · similarity 0.86)
- Recoverable Compression: A Multimodal Vision Token Recovery Mechanism Guided by Text Information (AAAI 2025 · similarity 0.84)
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.84)
- VIXEN: Visual Text Comparison Network for Image Difference Captioning (AAAI 2024 · similarity 0.84)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.84)
- Beyond Text: Frozen Large Language Models in Visual Signal Comprehension (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.
