Thermal Camera Detection

This repository is based on NVIDIA's tf_trt_models and jkjung's fork of NVIDIA's repository, along with the TensorRT repository. Below, you will find an index of the contents of this repository so you can use them based on your needs. However, to fully understand the pipeline, we advice you to follow them in order. Also, be sure to first follow the setup process for the scripts to work properly.

Setup

In order to use this repository, you must first make sure that the host computer has the latest NVIDIA drivers installed. Afterwards, you will need Anaconda3 to a virtual environment (conda) using the yml file provided in the repository.

Refer to these links to install the prerequisites:

Once these prerequisites have been met, clone this repository

$ git clone https://github.com/Hrayo712/tf_trt_models

Note: The repository includes tensorflow object detection API submodules under the third_party folder. These can be cloned automatically by adding the --recursive flag to the clone command. Otherwise, the install script will do it for you.

After cloning the repository, setup the conda environment with the given yml file to create the virtual environment with all the required dependencies. This process can take a while.

$ cd tf_trt_models
$ conda env create -f environment/tf1_12_gpu.yml

Note: Be sure to restart the terminal before attempting to create the new conda environment. This is to allow conda to initialize the base environment via the .bashrc file on the home directory

Afterwards, run the installation script. This bash script will update the submodules, pulling the tensorflow models from a particular commit, and install the object detection API to facilitate training and development.

$ conda activate tf1_12_gpu

$ ./install.sh

If everything goes smoothly, you should be able to seamlessly execute the scripts provided in this repository.

$ cd tensorflow_training

$ python model_builder_test.py

This script verifies the installation of the setup, along with its dependencies. You should see the following output.

..................
----------------------------------------------------------------------
Ran 18 tests in 0.062s

OK