Baidu's "Unlimited OCR" processes dozens of document pages in one pass by treating memory like human forgetting
Baidu has introduced an optical character recognition system capable of processing dozens of document pages in a single pass, significantly exceeding the ten-page limit typical of previous models. By utilizing a modified attention mechanism that maintains constant memory usage, the system avoids performance degradation as document length increases. This development currently sets a new standard on primary industry benchmarks for document processing efficiency.
Covered by 1 source
- TThe Decoder↗Jonathan KemperJul 5