@@ -23,22 +23,22 @@ Making the data collecting experience friendlier for the user has been one of ou
### Data Collecting
The jouney starts with data collection. Data plays a fundumental role in creating a great model, thats both acurrate and has great performance. The data collection program has been designed to generate data suitable for training the model. Added to that, it has been made to help our users to create an accurate model. With breaks betwen training sessions and intutive design, anyone can create a good training dataset.
The journey starts with data collection. Data plays a fundamental role in creating a great model, that's both accurate and has great performance. The data collection program has been designed to generate data suitable for training the model. Added to that, it has been made to help our users to create an accurate model. With breaks between training sessions and intuitive design, anyone can create a good training dataset.
Another feature of SignTrack Data Collect is flipping the image to generate data as if you signed with the other hand too! Creating a model that has the ability to make equaly accurate predictions on both hands.
Another feature of SignTrack Data Collect is flipping the image to generate data as if you signed with the other hand too! Creating a model that can make equally accurate predictions on both hands.
![](Assets/readme/Collecting1.gif"SignTrack Data Collecting Example 1")![](Assets/readme/Collecting2.gif"SignTrack Data Collecting Example 2")
### Model Train
Training a neural network can be confusing. From setting training parameters, such as activation functions and the number of neurons on each layer, to the model architecture to the number of epochs, everything in SignTrack is automated, with the power of AutoTrain, requiring minimal, or no adjustment from the user while training.
AutoTrain makes sure that you get the most out of your dataset. It sets all the training parameters for you, while automatically saving the best-performing model.
Training a neural network can be confusing. From setting training parameters, such as activation functions and the number of neurons on each layer, to the model architecture to the number of epochs, everything in SignTrack is automated, with the power of AutoTrain, requiring minimal or no adjustment from the user while training.
AutoTrain makes sure that you get the most out of your dataset. It sets all the training parameters for you while automatically saving the best-performing model.
### Sign Track Main
Utilizing the created model turned out to be an equally fundumental part of the project. Some people sign faster than the 24 frames that the model requires for making predictions. FastTrack is built into the __SignTrack main.py__ to solve this problem. It randomly duplicates the frames that the model's predictions are based on, while also making sure that the model makes predictions only on frames which the hands are on the frame, enhancing once again the overal performance.
Utilizing the created model turned out to be an equally fundamental part of the project. Some people sign faster than the 24 frames that the model requires for making predictions. FastTrack is built into the __SignTrack main.py__ to solve this problem. It randomly duplicates the frames that the model's predictions are based on, until it gets the desired amount of frames. At the same time, making sure that the model makes predictions only on frames which the hands are on the frame, enhancing once again the overall performance.
The consistent, unique and identifiable design continues on the main program, while keeping on the display the needed information to understand how the model performs on specific signs. That can save a lot of time for those who work on making their own SignTrack model.
The consistent, unique, and identifiable design continues on the main program while keeping on the display the needed information to understand how the model performs on specific signs. That can save a lot of time for those who work on making their SignTrack model.
![](Assets/readme/Detect1.gif"SignTrack Main Program Example 1")![](Assets/readme/Detect2.gif"SignTrack Main Program Example 2")
...
...
@@ -101,7 +101,7 @@ You can simply try to change the camera input selection in the first lines of co
* Find this line of code __cap = cv2.VideoCapture(0)__
* Change the default value zero to another one like 1 or 2
* Change the default value zero to another one, like 1 or 2
