-
Yue Chen authored
This tool is a gadget for offline probability training. A binary executable aom_entropy_optimizer will be generated in tools/. It parses a binary file consisting of counts written in the format of FRAME_COUNTS in entropymode.h, and computes the optimized probability table, which will be written to a new c file optimized_probs.c using the format in entropymode.c. Command line: ./aom_entropy_optimizer [directory of the count file] The input file can be either be generated from a single run by turning on entropy_stats experiment(counts are accumulated from frame to frame, and finally written to counts.stt), or be collected at a larger scale, at which a python script (will be provided soon) can be used to aggregate multiple stats output. Optimization for initial CDFs will be also supported later. Change-Id: I32070721aa8059439feb6b5a3a179f1001c66bb7
Yue Chen authoredThis tool is a gadget for offline probability training. A binary executable aom_entropy_optimizer will be generated in tools/. It parses a binary file consisting of counts written in the format of FRAME_COUNTS in entropymode.h, and computes the optimized probability table, which will be written to a new c file optimized_probs.c using the format in entropymode.c. Command line: ./aom_entropy_optimizer [directory of the count file] The input file can be either be generated from a single run by turning on entropy_stats experiment(counts are accumulated from frame to frame, and finally written to counts.stt), or be collected at a larger scale, at which a python script (will be provided soon) can be used to aggregate multiple stats output. Optimization for initial CDFs will be also supported later. Change-Id: I32070721aa8059439feb6b5a3a179f1001c66bb7