Research map / Architectures and efficiency / Attention and transformers
Aggregation Cross-Entropy for Sequence Recognition
Zecheng Xie, Yaoxiong Huang, Yuanzhi Zhu, Lianwen Jin, Yuliang Liu, Lele Xie
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 |
|---|---|---|---|
| replace regression-based ACE with cross-entropy-based ACE (ablation) | ACE (1D) with CRNN · IIIT5K | accuracy: 19.4 -> 82.3 | Improved |
| replace regression-based ACE with cross-entropy-based ACE (ablation) | ACE (1D) with CRNN · SVT | accuracy: 6.6 -> 82.6 | Improved |
| replace regression-based ACE with cross-entropy-based ACE (ablation) | ACE (1D) with CRNN · IC03 | accuracy: 12.0 -> 92.1 | Improved |
| replace regression-based ACE with cross-entropy-based ACE (ablation) | ACE (1D) with CRNN · IC13 | accuracy: 9.3 -> 89.7 | Improved |
| use ACE instead of the strongest listed baseline | ACE (2D) · SVT-P (50-word lexicon) | accuracy: 94.3 -> 94.9 | Improved |
| use ACE instead of the strongest listed baseline | ACE (2D) · CUTE | accuracy: 76.8 -> 82.6 | Improved |
| use ACE instead of the strongest listed baseline | ACE (2D) · SVT-P (no lexicon) | accuracy: 75.8 -> 70.1 | Hurt |
| use ACE with 2D prediction | scene text recognition · SVT-P (full lexicon) | accuracy: 83.7 -> 87.8 | Improved |
| use ACE instead of the strongest listed baseline | ACE (2D) · IC15 | accuracy: 68.2 -> 68.9 | Improved |
| use ACE instead of the strongest listed baseline | ACE (1D) · ICDAR 2013 HCTR (without LM) | CR: 87.43 -> 91.68 | Improved |
| use ACE instead of the strongest listed baseline | ACE (1D) · ICDAR 2013 HCTR (without LM) | AR: 88.79 -> 91.25 | Improved |
| use ACE instead of the strongest listed baseline | ACE (1D) · ICDAR 2013 HCTR (with LM) | CR: 96.32 -> 96.70 | Improved |
Closest papers by meaning
- EVALUATION OF NEURAL ARCHITECTURES TRAINED WITH SQUARE LOSS VS CROSS-ENTROPY IN CLASSIFICATION TASKS (ICLR 2021 · similarity 0.82)
- A Continuous Relaxation of Beam Search for End-to-End Training of Neural Sequence Models (AAAI 2018 · similarity 0.81)
- Trainable Decoding of Sets of Sequences for Neural Sequence Models (ICML 2019 · similarity 0.81)
- Handwritten Mathematical Expression Recognition via Attention Aggregation Based Bi-directional Mutual Learning (AAAI 2022 · similarity 0.81)
- Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification (NeurIPS 2026 · similarity 0.80)
- Cut Your Losses in Large-Vocabulary Language Models (ICLR 2025 · similarity 0.79)
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.
