End-to-End pipeline for car number plate recognition using Deep Learning
-
YoloV3 for License plate detection
-
Segmentation of digits using Image Pre-processing techniques
-
Trained CNN classifier for classifying segmented digits
-
Training pipeline for CRNN Optical Character Recognition model for predicting text on the detected plate (Model needs to be trained on more data to increase efficiency)
-
Tutorial notebooks
-
Flask application for inference using combination of YOLO followed by segmentation, and CNN classification
-
AWS Deeplens tutorial using pre-trained SSD model aws artifiacts
-
Airflow training pipelines
- CNN Training pipeline - CRNN Training pipeline
-
Dockerizing airflow pipelines
To fire up docker service, clone this repository and cd into this folder - docker-compose -f docker-compose-CeleryExecutor.yml up - docker-compose -f docker-compose-CeleryExecutor.yml up -d (detached mode)
Prior work on object detection repurposes classifiers to perform detection. Instead, object detection is framed as a regression problem to spatially separated bounding boxes and associated class probabilities.
A single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation. Since the whole detection pipeline is a single network, it can be optimized end-to-end directly on detection performance.
AWS DeepLens is a deep learning-enabled video camera. It is integrated with the Amazon Machine
Learning ecosystem and can perform local inference against deployed models provisioned from the AWS Cloud.
With AWS DeepLens, you train a project model using a supported deep learning modeling framework.
You can train the model on the AWS Cloud or elsewhere. Currently, AWS DeepLens supports Caffe, TensorFlow and Apache MXNet frameworks
https://docs.google.com/document/d/1Sa3jxZ_6bPCQz6wDBn-h-yO2jaeonfrUAkS2-ZWV7hU/edit?usp=sharing