AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Multi-Granularity Prediction with Learnable Fusion for Scene Text Recognition

Cheng Da, Peng Wang, Cong Yao

IJCV 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded scene text and structure · 12 citations (OpenAlex)

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 changedWhereMeasuredEffect
add Character A3 module (ablation)MGP-STR ViT · 6 STR benchmarks AVGaccuracy: 91.58 -> 92.73Improved
use 32x128 input with 4x4 patches instead of 224x224 with 16x16 (ablation)MGP-STR ViT · 6 STR benchmarks AVGaccuracy: 91.07 -> 91.58Improved
add BPE and WordPiece heads with CFS fusion (ablation)MGP-STR Vision · 6 STR benchmarks AVGaccuracy: 92.73 -> 93.35Improved
use Cumprod instead of Mean confidence score (ablation)MGP-STR CFS · 6 STR benchmarks AVGaccuracy: 93.28 -> 93.35Improved
add BPE head only (C+B) with CFS (ablation)MGP-STR Vision · 6 STR benchmarks AVGaccuracy: 92.73 -> 93.24Improved
add WordPiece head only (C+W) with CFS (ablation)MGP-STR Vision · 6 STR benchmarks AVGaccuracy: 92.73 -> 93.25Improved
replace explicit BCN language model with multi-granularity prediction (ablation)MGP-STR Vision · 6 STR benchmarks AVGaccuracy: 93.14 -> 93.35Improved
fuse heads with CFS on smaller DeiT-Tiny backbone (ablation)MGP-STR · 6 STR benchmarks AVGaccuracy: 89.36 -> 89.91Improved
fuse heads with CFS on DeiT-Small backbone (ablation)MGP-STR · 6 STR benchmarks AVGaccuracy: 91.01 -> 91.96Improved
swap DeiT init for MAE, DINO, DINOv2, CLIP or BLIP (ablation)MGP-STR CFS (200M iters) · 6 STR benchmarks AVGaccuracy: 93.70 vs 93.60, 93.47, 93.59, 93.60, 93.53No clear effect
replace CFS with Learnable Fusion Strategy (ablation)MGP-STR (200M iters) · 6 STR benchmarks AVGaccuracy: 93.70 -> 94.05Improved
increase LFS noised text from 8 to 256 (ablation)MGP-STR LFS · 6 STR benchmarks AVGaccuracy: 93.67 -> 94.05Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy