Slow Inference for the same model but different weights

[jupiter]

I have two set of IR files. XML files are identical (except some node names(postfix numbers) which makes sense) but BIN files are different. Here's my question, can weights that are more diverse affect inference performance on MyriadX? One model is running @ 27fps and the other 6fps. Same models with different weights. If so, why? I believe timing of layers should be the same for same models unless I'm missing something. TIA

[Iffa_Intel]

Hi,

Could you share your files if possible?

It would help us to further investigate the issue.

If you are talking about the weight format, definitely they would produce different result.

FP16 might perform faster but its accuracy might be less that FP32 weight format.

This might help you understand better.

Sincerely,

Iffa

[Iffa_Intel]

Greetings,

Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question. 

Sincerely,

Iffa