Complete research journey
Scope, observations, published evidence, six research tracks and open scientific questions.
Read journal →Exploratory technical research · Version 0.3 · 21 October 2026
Independent research notes · Literature and architecture
Understanding what comes next for Khmer OCR.
A living research journal on visual recognition, Khmer Unicode representation, modern CTC architectures, self-supervised pretraining, document understanding and offline efficiency.
Khmer OCRSVTRv2Self-supervised learningCTCResearch notesKhmer tokenization and CTC feasibility
What should one output token represent? Alignment constraints for visually composed Khmer script.
Read Study 01 →Anatomy of SVTRv2
Multi-Size Resizing, Feature Rearrangement, Semantic Guidance and published ablations.
Read Study 02 →Self-supervised visual learning
MAE, DiG, MaskOCR and LMIM: robust visual features without erasing Khmer character distinctions.
Read Study 03 →Evidence standard
Published benchmark evidence applies only to the studied corpora. Local observations are small and developmental; Khmer-specific transfer ideas remain hypotheses. No private government documents or evaluation receipts appear here.