Practical question regarding the engine file

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Say, I would be able to protect my SD or eMMC from being written, either by initramfs (which does not work yet) or maybe a fuse protection.

Then I would have to face the problem, that the “.engine” file needs to be created at least once. For practical reasons I cannot redirect this to RAM, since it would have to be re-created with every cold boot and would slow down the start. I would probably have to have an additional writable device, like an USB stick or so.

Or would it be possible to prepare an engine file once for all and deploy that to other machines, along with the scripts?

3

do you want to generate the engine once, put it some where in disk, and load it every time?
and, you deploy this engine file on other devices?

are the devices the same model of the device used to generate the engine? ==> if answer is yes, then you can.

4

do you want to generate the engine once, put it some where in disk, and load it every time?
and, you deploy this engine file on other devices?

This was the plan.

are the devices the same model of the device used to generate the engine? ==> if answer is yes, then you can.

Yes. Thanks

5

If you set “model-engine-file=” in the nvinfer config, it will generates the engine file.
And, with this config, if the engine exists, deepstream will load the engine instead of building it again.

does this meet your requirement?

6

Yes