TASK6_MLOPS_MASK-RCNN_MODEL_USED_FOR_INVOICE_CUTTING_OF_AUTOMOBILES
TASK DESCRIPTION:
Problem Statement: Create a project designed to solve the real use case, using either transfer learning example existing Mask-RCNN, VGG16, etc. or creating new model of Mask-RCNN, GANs, RNN, etc. to solve any real case problems or new problems.
Necessary requirements:
1. Make your own custom dataset using supervisely
2. Either create a new model or using existing model as transfer learning
3. Launch the training on aws cloud
Using this model we can create a powerful system in which we can directly do the process of invoice cutting or e-challan their will be no need for the traffic police to go and catch each and every person who jumps over the traffic signal or catch those who drive above the speed limits. I got this idea from the use of FASTags on the polling booth. Not only it can be used for the challan but it can also be used to find or monitor people. Number plates nowadays can be easily changed but what if we use a bar code or a unique number to assign every automobiles and they can not be changed easily. Using this method the work for the invoice cutting will be very easy for the traffic police department and also this can help them in several other things.
To perform this practical i wont be able to find a good quality of dataset as the fastag images are blur in most of the sites so i have used this concept for a simpler use in distinguishing car and bikes. When I will collect the complete dataset I will create a new model and write a separate blog for it.
For creating this project I have used supervisely to create and train the complete model.
Create a new workspace and upload the data you want to train on.And then use the bitmap to mark the object you seek in the image.
After that use the DSL code to do the data augmentation process.
After doing this choose the Neural Network type which you prefer building the entire system as you can use VGG16 , Resnet and many other neural networks. In my case I have used the Mask-RCNN model as it is the best model in present time.
Before training the model we need to provide a node to the supervisely Go to cluster and click on +ADD and it must have the Nvidia GPU support. In my case i have installed these drivers in my RHEL8 OS and that was not an easy part It took me almost 3–4 days just to create that setup as their were a lot of dependencies issue and driver settings we need to change in the system. I would recommend to use cloud platform for this process as it might take charge for atleast 1-2 hrs but your project will not consume this much extra time.
After adding the node you can easily train the model on the given dataset.
You can either download the weight or test the model on the supervisely.
I have provided a simple image as a test image and it gives me this output.\
I have created this project with my friend Aditya Sharma and we have decided to continue the project so as to solve the real used-case as I have explained above and for that the biggest constraint is the dataset. If anyone know about the fastag dataset than please letme know I have tried to search many times but didnt find anything useful.
I have done this part by creating a simple distinguishing program because i dont have any idea how much long that dataset process will take.
THANK YOU
LinkedIN profile: https://www.linkedin.com/in/akshat-soni-011b461a6