Browse categories
Explore
Fiverr Pro
English
$
USD
I will fix mediapipe GPU delegate errors on arm linux, docker, or headless
Richter
About this gig
Is MediaPipe GPU delegate failing on your ARM device, Docker container, or headless server?
Common errors I fix:
"Failed creating base context during opening of kernel driver"
"eglGetDisplay() returned EGL_NO_DISPLAY"
"Kernel module may not have been loaded"
GPU delegate silently falls back to CPU with no error
MediaPipe works on desktop but crashes on edge/embedded
I compiled MediaPipe 0.10.35 from Bazel source with EGL/GBM GPU delegate on ARM Mali GPU running fully headless (no X11, no Wayland, no Xvfb). Achieved 2.3x speedup over CPU.
What most sellers don't know:
MediaPipe GPU delegate uses EGL, NOT CUDA even on Jetson
EGL requires a display server by default I patched it to use GBM (Generic Buffer Management) for true headless
This works on Mali (RK3576/RK3588), VideoCore (RPi 5), and Adreno GPUs
Live demo (terminal recording): https://asciinema.org/a/Mv4LEGvaroBSs6oJ
I handle:
ARM aarch64 compilation from source (Bazel + CMake)
Docker GPU pass-through for MediaPipe
Headless EGL/GBM patching
Performance benchmarking (CPU vs GPU)
Platform: Python 3.10-3.12, Linux ARM64, Docker-compatible
Get to know Richter
Richter
- FromChina
- Member sinceOct 2024
- Last delivery1 year
Languages
English, Chinese, German
I build computer vision systems that ship — on NVIDIA CUDA servers and ARM edge.
Not demos. Production.
6 projects deployed in 12 months: YOLO detection + tracking on CUDA and NPU (17x speedup), multi-camera RTSP pipelines with FFmpeg hardware decoding, MediaPipe GPU compiled from source for ARM Mali (2.3x faster, headless), PyTorch custom model training, and rPPG contactless vital signs from video.
Stack: Python, C++, PyTorch, OpenCV, CUDA, ONNX, YOLO, Docker. GPUs: RTX 4060 Ti, Hailo-8L NPU, Mali-G52.
3600+ lines in a real school. 20K+ lines in a shipping edge AI product.
My Portfolio
FAQ
Does this work without a monitor connected?
Yes! That's the whole point. I patched MediaPipe to use EGL GBM platform instead of X11. Your device can run GPU inference in a server rack, Docker container, or embedded box with no display attached.
What hardware do you support?
Tested on RK3576 (Mali-G52 MC3). Also compatible with RK3588 (Mali-G610), Raspberry Pi 5 (VideoCore VII), and other ARM SBCs with EGL 1.5 support. If your GPU supports OpenGL ES 3.1+, it should work.
Do you also support NVIDIA Jetson?
Jetson uses NVIDIA GPU with EGL — the headless approach is similar but not yet tested in my setup. I can help with Jetson EGL debugging, but headless compilation has only been verified on Mali so far.
Will this fix my TFLite GPU delegate too?
Partially. MediaPipe and TFLite share the same underlying EGL/GLES infrastructure. The GBM headless approach applies to both. I can advise on TFLite-specific steps.
Can you work with my existing codebase?
Yes. Share your code via GitHub/GitLab or zip file. I'll diagnose the GPU delegate failure, apply the fix, and verify it runs.
