DavidTorresOcana / ros_object_detector

A ROS package to run Tensorflow models for object detection and image segmentation
6 stars 1 forks source link

Object trainig #2

Closed melikavhf closed 6 years ago

melikavhf commented 6 years ago

hello . how can I train my own object and how to replace in this package? how should I create my own model? Is there any package to create these files? sorry I'm not familiar with object detection .

DavidTorresOcana commented 6 years ago

Hello

I recommend you to explore/use Tensorflow's Object Detection libraries/code: https://github.com/tensorflow/models/tree/master/research/object_detection

With that, you should be able to train models on your data, freeze your model and include it into this ROS package. Once you have your model trained and frozen , place it inside /include/ros_object_detector and run the node with the appropriate parameters. You will need a Tensorflow frozen graph on .pb format and the labels definition into .pbtxt format.

Please explore /include/ros_object_detector to see how the "pothole_detector" files estructure is.

Object detection (with Convolutional Neural Networks in this case) is Machine Learning so, if you are not familiar with Machine Learning, I recommend you this course and take a look to Tensorflow tutorials (if you were not familiar with it)

Regards

melikavhf commented 6 years ago

Thanks for attention . how can I create .pbtxt file and .pb? and how should I connect these file to each other?

melikavhf commented 6 years ago

and do I need special dataset? or no ? just .pb ?

DavidTorresOcana commented 6 years ago

Hello

With all the respect, this is not related with any issue of the repo so so we need to closing it. Feel free to contact me personally on email. Please take a look to the references above to learn more about machine learning. If you were looking for a general object detector, Yolov2 is coming to this repo soon, so you may only need to wait.

Regards