First step of this project was dataset downloading second step was dividing three part as train, test and validation dataset images. %70 of the dataset was train data , %20 of dataset was validation data and %10 of dataset was test data. Pretrained Resnet-50 and VGG-16 models were trained for comparison purposes. Here is results :
-
Precision - Precision is the ratio of correctly predicted positive observations to the total predicted positive observations.
-
Recall (Sensitivity) - Recall is the ratio of correctly predicted positive observations to the all observations in actual class.
-
F1 score - F1 Score is the weighted average of Precision and Recall.
Classification Report
precision recall f1-score support
go 0.98 0.72 0.83 241
stop 0.83 0.99 0.90 335
accuracy 0.88 576
macro avg 0.90 0.85 0.86 576
weighted avg 0.89 0.88 0.87 576
RESNET-50 Accuracy and Loss Graphic
Classification Report
precision recall f1-score support
go 0.95 0.80 0.97 241
stop 0.90 0.97 0.92 335
accuracy 0.97 576
macro avg 0.94 0.82 0.98 576
weighted avg 0.90 0.90 0.98 576
VGG-16 Accuracy and Loss Graphic
VGG-16 gave better result so it's picked to contunie of project.
After ros2bag file these step below done.
-
mkdir foxy_ws && cd foxy_ws #create workspace -
mkdir src && cd src #create source -
git clone .... #clone darknet_ros repo -
source /opt/ros/foxy/setup.bash #Set up your environment by sourcing the following file. -
rosdep install -i --from-path src --rosdistro foxy -y #Install dependencies from src -
sudo apt-get install ros-foxy-vision-msgs -
source install/setup.bash #This script extends the environment with the environment of other prefix paths which were sourced when this file was generated as well as all packages contained in this prefix path. -
colcon build #makes file as ros file -
ros2 bag play ros_bag_file_name.bag #plays video in rosbag file
-
source /opt/ros/foxy/setup.bash -
source install/setup.bash -
ros2 launch darknet_ros yolov3.launch.py
-
source /opt/ros/foxy/setup.bash -
source install/setup.bash -
ros2 topic list # to see published nodes inside packages
-
ros2 topic echo /darknet_ros/bounding_boxes # to see message inside of the node
The call back functions of the subscriber nodes I created were not working because they could not synchronize.After searching the internet. As a method, I learned that the metadata in the header, the time stamp, should be fixed. For this, I fixed the time of the raw images in the classifier node and gave them to the message service.Then I produced results by inserting images into the model prediction and on a new publisher node. I published the name of classes and the scores of them. Here is the result after solving simultaneous synchronization problem :
To train models :
- https://www.kaggle.com/suniliitb96/tutorial-keras-transfer-learning-with-resnet50/notebook
- https://www.kaggle.com/kmkarakaya/transfer-learning-for-image-classification/notebook
- https://github.com/pytorch/vision/tree/main/torchvision
- https://github.com/WZMIAOMIAO/deep-learning-for-image-processing/
To calculate presicion ,recall metrics :
- https://blog.exsilio.com/all/accuracy-precision-recall-f1-score-interpretation-of-performance-measures
- https://www.kaggle.com/surajyathinatti/transfer-learning-confusion-matrix-class-report/data
To create publisher and subscriber nodes :
- https://docs.ros.org/en/foxy/Tutorials/Writing-A-Simple-Py-Publisher-And-Subscriber.html
- https://automaticaddison.com/create-a-basic-publisher-and-subscriber-python-ros2-foxy/
To objects classification :
To synchronization problem solving :