Failed to build TensorRT engine

Omniverse Apps Audio2Face
1

Hi everybody,

my system failed to build the TensorRTEngine.

System:
Win10 19044 (relatively clean install)
AMD Ryzen 7 CPU
RTX 3070
16gb RAM
Nvidia 516,94
Edit:
Audio2Face version 2022.1.1

System runs in virtual machine with direct PCI passthrough to the RTX card.
through. AI Models were already used on that system.
Log is attached.

kit_20220910_155145.log (689.6 KB)

1 Like
2

I have the same problem.
kit_20221031_195837.log (3.1 MB)
Windows 11

|Processor|12th Gen Intel(R) Core™ i9-12900HK 2.50 GHz|

|System type|64-bit operating system, x64-based processor|

NVIDIA GeForce RTX 3050 Ti Laptop GPU

1 Like
3

kit_20221123_225808.log (1.3 MB)

The same.
Failed on my laptop.
i9-12900H 32GB RAM RTX3050
Windows 11 Pro 22H2 (clean installed)

However, it’s ok on my desktop PC.

4

I think this issue appears only on laptops, I have a laptop and also faiced the TensorRT failed to build engin issue as well as my brother’s laptop.

Maybe they can fix this issue on the next update and make it possible for laptop users to run this program without errors.

5

After reviewing logs, I found it may cause by VRAM shortage. During TensorRT building, it almost take up all VRAM.

6

It means there is no way to build that TensorRT engine on a 4 gb videocarded laptop like mine?

However I have successfully build the TensorRT engine for emotion tab, but can’t do it on auto-emotion, all I need is just that emotion tab, I’m not in a big need of auto-emotion, maybe there is a way to make work only with one emotion tab?

And also I can successfully build the old no eyes version’s TensorRT, I don’t have problems there, maybe there is a way somehow to add that emotion tab on the no eyes old version ? Cus old one don’t have emotions.

2022-11-26_16-30-06
2022-11-26_16-30-061919×1003 220 KB

7

Hi!
Experiencing the same problem. Did you manage to resolve it somehow?

upd: okay, I figured it out.

  1. Create new scene
  2. Go Audio2Face → +Head Template
  3. Audio2Face → +A2F Core
  4. Change Attached prism dropdown to whatever is there.
8

Does it for the no eyes version? No eyes version has lack of emotions and I’ve succesfully created TensorRT on that no eyes version, but I need to create that TensorRT for the Full-face

9

Works with no eyes version, yes. RIght now tried on the eyes version and the engine builds up succesfully.

10

So the workflow is:

  1. File → New
  2. Audio2Face → Head Template → Male Head Full Face
  3. Audio2Face → A2F Core → Full Face Core.
  4. In Audio2Face window scroll all the way down and add the Audio Player → Regular Audio Player
  5. On the upper panel go In Audio2Face → Rebuild TenzorRT → Full face and wait. You’re done.
1 Like
11

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.