Search papers, labs, and topics across Lattice.
2
0
4
1
Fine-tuning an existing OCR model, even if not pre-trained on target languages, surprisingly beats end-to-end training for production-scale multilingual OCR in the Indian context, offering a sweet spot of accuracy and speed.
Current VLMs stumble significantly when faced with the cultural and linguistic nuances of the Indian subcontinent, as revealed by the new IndicVisionBench benchmark.
