Tensorflow-Carla-Object-Detection

With this repo you can train your own object detection classifier with the tensorflow object detection api. I use it to detect several objects in the Carla simulator. For this purpose I created my own dataset which can be downloaded from my other github repository. This is tested for Ubuntu 16.04 but should also work under windows and other linux distributions.

Youtube demonstration video of the trained classifier:

A more detailed instruction on how to train this object detection classifier can be found under: Link

Basic steps to follow:

1. Install Tensorflow GPU support

   Go to the Tensorflow website and follow the step described here. You will also need to install CUDA and cuDNN which is also described on the website.

2. Download Tensorflow object detection api

   A detailed description can be found here.

3. Download this repository

   Clone this repository into the ./tensorflow/models/research/object_detection/ folder.

4. Download the a pretrained model from TensorFlow's model zoo

   To train the model you will need to use a pretrained model, otherwise training would consume to much time. I my case I used the Faster-RCNN-Inception-V2-COCO model and downloaded it from Tensorflow's model zoo. Extract the pretrained model into the ./tensorflow/models/research/object_detection/ folder.

5. Download dataset and copy files

   Download the dataset from my other github repo and extract the images into the "images" of this repository e.g. ./tensorflow/models/research/object_detection/Carla_object_detection/images/

6. Generate training data

7. Configure training

8. Run training

9. Export inference graph

10. Use the trained classifier