Using gpu on server, tensorflow object detection api installation

TensorFlow General Discussion
,
1

HI TEAM,
I am using vs-code and connecting to a server using remote window. the server has GPU (CUDA), and i am trying to install tensorflow object detection API. the remote window connecting to server is linux based, although my laptop has Windows.
Coming to the issues, i started with creating virtual environment and activating it. I am using Python 3.10 version. i installed tensorflow library 2.13.0 version, and i tested it by running the below command (also showing the output):

(.tfod310) admaryad@ubuntuserver1:~/SSD$ python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type “help”, “copyright”, “credits” or “license” for more information.

import tensorflow as tf
2024-06-13 09:54:09.173880: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-06-13 09:54:09.739921: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

What does it mean, how should i make it run smoothly, what changes needs to be done?

Please help. Thank you.

Regards

1 Like
3

None. Just use the test commands to check it detects your gpu. In your Python code after importing the modules, add:

tf.config.list_physical_devices('GPU')

You may need to print it if it doesn’t automatically, I dont remember right now.

The warnings just tell you that you could re-build tensorflow to be optimised to use certain CPU instructions, and that it didn’t find TensorRT which is an NVIDIAs API for fast inference.

5

Official TensorFlow 2.16 + Python 3.12 – JARaaS Hybrid RAG - 6/17/2024
Note: Sources at the end of the response

To install and run TensorFlow Object Detection API on a remote server with a GPU, you need to ensure that both TensorFlow and the necessary GPU drivers and libraries (CUDA, cuDNN) are properly installed. Here are the steps you should follow:

1. Install GPU Drivers

Ensure that the NVIDIA GPU drivers are installed on your server. You can check this by running:

nvidia-smi

If the drivers are not installed, follow the NVIDIA driver installation guide.

2. Install CUDA and cuDNN

Install the CUDA toolkit and cuDNN library. You can use conda or download them from NVIDIA’s website.

Using conda:

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0

3. Install TensorFlow with GPU Support

Make sure you install the GPU version of TensorFlow. Since you’re using Python 3.10, install TensorFlow 2.13.0 as follows:

pip install tensorflow==2.13.0

4. Setting Up the Object Detection API

Clone the TensorFlow models repository and install the Object Detection API.

Clone the repository:

git clone https://github.com/tensorflow/models.git
cd models/research

Install dependencies:

# Protocol Buffers compilation
protoc object_detection/protos/*.proto --python_out=.

# Install the Object Detection API
pip install .

5. Verify GPU Configuration

Run a simple TensorFlow GPU check script to ensure TensorFlow can access the GPU:

import tensorflow as tf
print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU')))

The output should show the number of GPUs available.

Troubleshooting TF-TRT Warning

The warning you are seeing about TF-TRT (TensorRT) means TensorRT is not installed but is required if you need TensorRT optimizations. Installing TensorRT is optional unless you need specific optimizations. Here’s how you can install it if needed:

Using conda:

conda install -c anaconda tensorrt

Sample Execution

Ensure your virtual environment is active, and run a basic example to test if the Object Detection API is set up correctly:

python object_detection/builders/model_builder_tf2_test.py

This will test your setup for the Object Detection API.

Summary

Make sure all dependencies are correctly installed, and your TensorFlow environment is set to utilize the GPU. If TensorRT is not required for your current assignments, the warning can be ignored unless you decide to use TensorRT for performance improvements.

Below is further documentation for different aspects of the setup:

Sources:

  • GPU Support Documentation: distribution.ipynb (internal document)
  • Setting up TensorFlow with GPU on a remote server: pip.md (internal document)
  • Object Detection API setup: dtensor_keras_tutorial.ipynb (internal document)

If you need further assistance, please refer to your internal guides or documentation for specific instructions tailored to your environment.