I can’t find a way to use Tensorflow Object Detection API without breaking CUDA support on Windows 11. Tried every single tutorial I could find on the internet, but still nothing works.

I’ve tried searching everywhere but couldn’t find anything that helped. All the answers seem to be outdated and no longer work. I even tried ChatGPT 4o, Gemini Advanced and Perplexity AI without success.

I’m using Tensorflow with GPU/CUDA support on Windows. It’s working fine, but for some pre-trained models I need Object Detection API, and this is where everything breaks. I tried just installing through pip, but it updates my Tensorflow beyond version 2.10, which breaks CUDA support. I’ve tried installing without dependencies, but then it doesn’t work. I’ve tried using protobuf to compile it, but if I set it to use TF 2.10, it doesn’t work. I’ve tried many different ways of installing it, but it always either updates tensorflow or just doesn’t work at all.

I’ve tried configuring tensorflow with GPU support via Docker, WSL2, tried straight in a Linux Mint 21.3 in dual boot, but none of these even work with GPU support, doesn’t matter how many tutorials I follow. The closest I got was using WSL2 Ubuntu, which recognized my CUDA installation according to nvidia-smi, but didn’t work with tensorflow.

As a last desperate attempt I tried using Google Colab, but despite having my image dataset on my Google Drive, and even buying credits to use better GPUs, for some reason it always fails to find some of my images and the execution halts.

I’m desperate, I need to make this work because I need to deliver my master’s thesis, and I’m running out of time (really, last week to deliver).

Is there any way to use Object Detection API on windows with my already working setup? (Windows 11 with latest Nvidia drivers, RTX 3070, Python 3.9, Tensorflow 2.10, CUDA 11.2 and cuDNN 8.1). I’m going to sleep now, but any help is welcome. If needed, I can give more details when I wake up.