I am exploring ways to improve the performance(single-thread / multi-thread) of tesseract-ocr inference. Performance profiling tess-ocr inference (CLI) using Intel Vtune <https://www.intel.com/content/www/us/en/developer/tools/oneapi/vtune-profiler.html> showed very little usage of the CPU's vector registers (AVX-2 and 512). The default inference using tesseract works on a single image (batch_size=1). I think batched inference would drive better utilization of the vector registers and thus improve the inference throughput of tess-ocr. Is there a way to enable batched inference on tess-ocr ?
Thanks in advance Regards Vishnu other details: - system : Intel Xeon 8380 (Icelake), Ubuntu 22.04(5.15), GCC 11.3 - tesseract : built from source (5.3.0), tessdata - configure flags : --enable-float32 --disable-opencl --disable-graphics 'CXXFLAGS=-O3 -mavx512f -mfma' other performance observations: - considerable openmp pause times observed for the multi threaded runs - LSTM --> Tanh lookup times are considerably high (~ 50ms per lookup) -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/982ae47a-6b2f-4565-9e82-0aadf1f09d65n%40googlegroups.com.
