Research map / Architectures and efficiency / CNN design
Recursive Recurrent Nets With Attention Modeling for OCR in the Wild
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 JOINT with R²AM | lexicon-free OCR · SVT | accuracy: 71.7 -> 80.7 | Helped |
| Replace JOINT with R²AM | lexicon-free OCR · ICDAR13 | accuracy: 81.8 -> 90.0 | Helped |
| Add recursive CNN with 3 iterations (ablation) | Base CNN · Synth90k, SVT, ICDAR13 | accuracy: 91.9 -> 94.2; 75.1 -> 78.9; 85.7 -> 88.5 | Helped |
| Increase recursive CNN iterations from 2 to 3 (ablation) | Recursive CNN · Synth90k, SVT, ICDAR13 | accuracy: 93.3 -> 94.2; 77.1 -> 78.9; 87.3 -> 88.5 | Helped |
| Use recursive instead of recurrent CNN with 3 iterations (ablation) | CNN image encoder · Synth90k, SVT, ICDAR13 | accuracy: 93.5 -> 94.2; 76.9 -> 78.9; 87.4 -> 88.5 | Helped |
| Add an RNN language model (ablation) | Base CNN · Synth90k, SVT, ICDAR13 | accuracy: 91.9 -> 93.4; 75.1 -> 76.2; 85.7 -> 86.4 | Helped |
| Feed image features to the RNN at every time step (ablation) | Base CNN + RNN_1c · Synth90k, SVT, ICDAR13 | accuracy: 93.4 -> 93.5; 76.2 -> 76.9; 86.4 -> 87.2 | Helped |
| Use a two-layer unfactored RNN instead of a one-layer unfactored RNN (ablation) | Base CNN + RNN_1u · Synth90k, SVT, ICDAR13 | accuracy: 93.5 -> 93.7; 76.9 -> 77.9; 87.2 -> 87.6 | Helped |
| Factor the two-layer RNN, restricting image features to the second stack (ablation) | Base CNN + RNN_2u · Synth90k, SVT, ICDAR13 | accuracy: 93.7 -> 94.0; 77.9 -> 78.8; 87.6 -> 88.0 | Helped |
| Add soft attention to the factored RNN (ablation) | Base CNN + RNN_2f · Synth90k, SVT, ICDAR13 | accuracy: 94.0 -> 94.3; 78.8 -> 79.1; 88.0 -> 88.9 | Helped |
| Add backward or bidirectional RNNs (ablation) | character-level language model · scene text recognition | accuracy | No clear effect |
Closest papers by meaning
- Gated Recurrent Convolution Neural Network for OCR (NeurIPS 2017 · similarity 0.91)
- Reading Scene Text in Deep Convolutional Sequences (AAAI 2016 · similarity 0.89)
- Char-Net: A Character-Aware Neural Network for Distorted Scene Text Recognition (AAAI 2018 · similarity 0.87)
- Making Convolutional Networks Recurrent for Visual Sequence Learning (CVPR 2018 · similarity 0.87)
- Look Closer to See Better: Recurrent Attention Convolutional Neural Network for Fine-Grained Image Recognition (CVPR 2017 · similarity 0.86)
- CLIP as RNN: Segment Countless Visual Concepts without Training Endeavor (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.