Issue background:
sdk version: a modified version based on ti-processor-sdk-rtos-j721s2-evm-08_06_00_11
I have the structure of Model A with conv3*3, relu and upsample. The inference speed of this model varies greatly among different versions after quantifying with different SDKs, as shown below
1. Using Model A (SDK: 8.2) for single frame inference on TDA4VM version takes 26ms
2. Using Model A (SDK: 8.4_eco) for single frame inference on TDA4ECO version takes 32ms
3. Using Model A (SDK: 8.5_eco) for single frame inference on TDA4ECO version takes 28 ms
4. Using Model A (SDK: 8.6) for single frame inference on TDA4AL version takes 42.5ms
Who meet this quesetions? I hope for your help, thanks!
