TI中文支持网Part Number:AM62A7
TIDL version 09.02.06
Using int16 quantization model, the effect is acceptable, but using int8 quantization, the effect is very poor. How to locate the difference caused by which layer?
Found a problem: need to remove the mul, sub, and mul in Figure 1 (marked with yellow boxes), int16 quantization is normal, and the modified model output is shown in Figure 2
Vivian Gao：
您好
请等待英文论坛资深工程师的回复，后续问题可以继续在英文帖子中交流。
