By Samay Pashine, Lan Nguyen, Yuyao Guo and Zhangchen Sun
Object Detection with Region-Based Convolution Neural Network based on VGG Backbone is the project that we built R-CNN model using python and deep learning libraries like tensorflow, scikit-learn with airplane dataset.
Object detection is a computer vision technique that allows us to identify and locate objects in an image or video.
Our vision aims to utilize the R-CNN, which uses bounding boxes across the object regions to evaluate convolutional networks independently on all the region proposals and in turn classify multiple image regions into the proposed class. We further extend R-CNN by leveraging self-attention mechanism to enable the model to focus on discriminative parts on region proposals. We have faith that our application will advance the object detection accuracy through large-scale screening.
Please follow this link to download our used dataset.
- Read the source and do some research to understand more about the dataset and its topic.
- Study to build R-CNN model in python from scratch.
- Study about the OpenCV functionalities
- Perform experiments with and without Attention Mechanism on the R-CNN model based VGG backbone.
- Analyze the accuracy rate more deeply and extract insights.
- Visualize analysis on Model Loss graphs.
- Python and some neccessary libraries such Tensorflow, keras, OpenCV, scikit-learn.
- Visual Studio Code to train models.
- Building R-CNN model
The architecture is built by Tensorflow. More details can be found in train.py.
self.base_model = VGG16(weights='imagenet', include_top=True)
self.freeze_layer(base_layer_train_flag)
print("\n[INFO]. VGG-16 Model Summary : ")
self.base_model.summary()
X = self.base_model.layers[-2].output
self.head = Dense(2, activation="softmax")(X)
self.model = Model(self.base_model.input, self.head)Our model has the accuracy of 97.61 % for the train dataset and 92.41 % for the train dataset with attention layer in the model.
We successfully built a deep neural network model by implementing Region-Based Convolutional Neural Network (R-CNN) to automatically detect airplane image with high accuracy rate 97.61 %. In addition, we also add Attention Mechanism in our experiment for further advancement.