AI Paper Atlas

Research map / Image generation / Editing and style transfer

SimAN: Exploring Self-Supervised Representation Learning of Scene Text via Similarity-Aware Normalization

Canjie Luo, Lianwen Jin, Jingdong Chen

CVPR 2022 · Image generation › Editing and style transfer · approach: Image-to-image translation · 41 citations (OpenAlex)

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
Use similarity-based local style alignment (SimAN) instead of global-statistics AdaIN (ablation)ResNet-FCN-Att. probe · IIIT5Kaccuracy: 9.7 -> 22.2Improved
Use similarity-based local style alignment (SimAN) instead of global-statistics AdaIN (ablation)ResNet-FCN-Att. probe · IC03accuracy: 9.1 -> 22.3Improved
Use similarity-based local style alignment (SimAN) instead of global-statistics AdaIN (ablation)ResNet-FCN-Att. probe · IC13accuracy: 11.1 -> 24.1Improved
Add adversarial training (ablation)SimAN representation learner · Scene-text probe (IIIT5K, IC03, IC13 average)accuracy (%): 68.8 -> 69.0Improved
Add adversarial training (ablation)SimAN image generator · Generated image distributionFID: 24.0 -> 23.2Improved
Add self-supervised pretraining on unlabeled SynthTextSimAN recognizer with 10K labeled SynthText · IIIT5Kword accuracy (%): 35.0 -> 41.1Improved
Add self-supervised pretraining on unlabeled SynthTextSimAN recognizer with 10K labeled SynthText · IC03word accuracy (%): 37.6 -> 42.9Improved
Add self-supervised pretraining on unlabeled SynthTextSimAN recognizer with 10K labeled SynthText · IC13word accuracy (%): 38.6 -> 43.9Improved
Pretrain the encoder on Real-300K rather than SynthTextSimAN frozen-backbone probe with SynthText-trained probe · IIIT5Kaccuracy: 60.8 -> 68.9Improved
Add a sequence model (2*RNN) to the self-supervised encoder (ablation)SimAN CTC probe · IIIT5Kaccuracy: 1.5 -> 0.0Hurt
Add a sequence model (2*RNN) to the self-supervised encoder (ablation)SimAN attention probe · IIIT5Kaccuracy: 22.2 -> 6.4Hurt
Use two BiLSTMs rather than one in the probe (ablation)SimAN CTC probe · IIIT5Kaccuracy: 57.4 -> 60.8Improved

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