AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation

Sharon Fogel, Hadar Averbuch-Elor, Sarel Cohen, Shai Mazor, Roee Litman

CVPR 2020 · Label-efficient and robust learning › Semi-supervised learning · approach: GAN-based semi-supervised learning · 166 citations (OpenAlex)

This paper tried a common change: Add batch / layer normalization

Add conditional batch normalization to the generator · RIMES · WER: 12.46 \pm 0.30 -> 11.68\pm 0.29 Improved

What 431 accepted papers report for this change

See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.

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 100k ScrabbleGAN samples and fine-tune on original dataHTR framework · RIMESWER: 12.29\pm 0.15 -> 11.32\pm 0.31Improved
Add 100k ScrabbleGAN samples and fine-tune on original dataHTR framework · IAMWER: 25.10\pm 0.49 -> 23.61\pm 0.36Improved
Generate 100k samples with CVL style and IAM lexiconHTR trained on IAM · CVL testWER: 39.95\pm 0.91 -> 40.24\pm 0.51No clear effect
Use the CVL lexicon instead of the IAM lexicon in generated training samplesHTR trained on IAM · CVL testWER: 39.95\pm 0.91 -> 35.98\pm 0.38Improved
Generate samples with both CVL style and CVL lexiconHTR trained on IAM · CVL testWER: 35.98\pm 0.38 -> 29.75\pm 0.67Improved
Generate samples with both CVL style and CVL lexiconHTR trained on IAM · CVL testNED: 14.52\pm 0.51 vs 15.62\pm 0.15Improved
Use the proposed gradient balancing with alpha 1 (ablation)ScrabbleGAN-trained HTR · RIMESWER: 12.64\pm 0.20 -> 11.68\pm 0.29Improved
Add gradient balancing rather than using no gradient balancing (ablation)ScrabbleGAN-trained HTR · RIMESWER: 12.64\pm 0.20 -> 12.83\pm 0.28Hurt
Set proposed gradient-balancing alpha to 0.1 instead of 1 (ablation)ScrabbleGAN-trained HTR · RIMESWER: 11.68\pm 0.29 -> 12.28\pm 0.49Hurt
Set proposed gradient-balancing alpha to 10 instead of 1 (ablation)ScrabbleGAN-trained HTR · RIMESWER: 11.68\pm 0.29 -> 12.03\pm 0.27Hurt
Remove the LSTM from the CNN recognizer (ablation)ScrabbleGAN · RIMESWER: 13.80\pm 0.30 -> 11.68\pm 0.29Improved
Use a CNN recognizer without a recurrent head instead of a CRNN recognizer (ablation)ScrabbleGAN · RIMESWER: 12.18\pm 0.24 -> 11.68\pm 0.29Improved

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