Data Segmentation

TensorFlow provides most promising techniques for semantic image segmentation with Deep Learning known as DeepLab,The aim is to assign semantic labels (e.g. car, human, cat) to every pixel in the input image.

Semantic image segmentation predicts whether each pixel of an image is assigned with a particular classsuch that pixels with the same label share certain characteristics. On the other hand object detection, which detects objects in rectangular regions, and image classification, which classifies the overall image.

The current implementation includes the following features:

 DeepLab-v3+ implemented on top of TensorFlow. This includes DeepLab-v3+ models built on convolutional neural network (CNN) as backend architecture for the most accurate outputs, Used mainly for server-side deployment.

The model will create a mask over the target objects with high accuracy.

Much advancement happened once after the first introduction of DeepLab model. Some of them are given below:

– Object Scale Modeling

– assimilation of contextual information

– improvised training procedures

– CNN feature extractors

From the DeepLab-V3+ (which is extension of third Version DeepLab-v3), it was added with effective decoder module which will refine the segmention results. if we apply  deep convolution  to both atrous spatial pyramid pooling and decoder modules, resulting in stronger encoder and decoder network..