Hello all,
did anyone manage to build mediapipe wheel with GPU linkage for python3.10 on JP6.2 (CUDA12.6)? Do you know if there are any repositories with pre-built wheels for AI/ML stuff for Jetson platform?
I am trying to find one too…
Hi,
Unfortunately, they didn’t support Jetson officially:
opened 02:13PM - 25 Jun 24 UTC
closed 01:50AM - 11 Jul 24 UTC
type:build/install
task::all
platform:embedded-linux-arm
### OS Platform and Distribution
Ubuntu 22.04
### Compiler version
_No respon… se_
### Programming Language and version
Python 3.8 / 3.10
### Installed using virtualenv? pip? Conda?(if python)
Built from source, generating Python wheel and attempt to pip install wheel
### MediaPipe version
0.8.11
### Bazel version
Bazelisk v1.8.1
### XCode and Tulsi versions (if iOS)
_No response_
### Android SDK and NDK versions (if android)
_No response_
### Android AAR (if android)
None
### OpenCV version (if running on desktop)
4.9.0
### Describe the problem
I attempted to build the Mediapipe package for ARM64 Linux by following the following walkthrough (https://www.forecr.io/blogs/ai-algorithms/how-to-download-build-mediapipe-on-nvidia-jetson-xavier-nx), which worked on Jetpack 5.1.3. This, however, only let me use the CPU as delegate with TFLite and I wanted to try on Jetpack 6.0.2. The wheels compile but the installation of the wheel fails with this very short error.
### Complete Logs
```shell
idealab@ubuntu:~/mediapipe/dist$ python3 -m pip install mediapipe-dev-cp38-cp38-linux_aarch64.whl
ERROR: Invalid requirement: 'mediapipe==dev': Expected end or semicolon (after name and no valid version specifier)
mediapipe==dev
^
```
Maybe other users can share their experiences.
Thanks.