1# Segmentation 2 3<img src="../images/segmentation.png" class="attempt-right" /> 4 5## Get started 6 7_DeepLab_ is a state-of-art deep learning model for semantic image segmentation, 8where the goal is to assign semantic labels (e.g. person, dog, cat) to every 9pixel in the input image. 10 11<a class="button button-primary" href="https://storage.googleapis.com/download.tensorflow.org/models/tflite/gpu/deeplabv3_257_mv_gpu.tflite">Download 12starter model</a> 13 14## How it works 15 16Semantic image segmentation predicts whether each pixel of an image is 17associated with a certain class. This is in contrast to 18<a href="../object_detection/overview.md">object detection</a>, which detects 19objects in rectangular regions, and 20<a href="../image_classification/overview.md">image classification</a>, which 21classifies the overall image. 22 23The current implementation includes the following features: 24<ol> 25 <li>DeepLabv1: We use atrous convolution to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks.</li> 26 <li>DeepLabv2: We use atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales with filters at multiple sampling rates and effective fields-of-views.</li> 27 <li>DeepLabv3: We augment the ASPP module with image-level feature [5, 6] to capture longer range information. We also include batch normalization [7] parameters to facilitate the training. In particular, we applying atrous convolution to extract output features at different output strides during training and evaluation, which efficiently enables training BN at output stride = 16 and attains a high performance at output stride = 8 during evaluation.</li> 28 <li>DeepLabv3+: We extend DeepLabv3 to include a simple yet effective decoder module to refine the segmentation results especially along object boundaries. Furthermore, in this encoder-decoder structure one can arbitrarily control the resolution of extracted encoder features by atrous convolution to trade-off precision and runtime.</li> 29</ol> 30 31## Example output 32 33The model will create a mask over the target objects with high accuracy. 34 35<img alt="Animation showing image segmentation" src="images/segmentation.gif" /> 36 37## Read more about segmentation 38 39<ul> 40 <li><a href="https://ai.googleblog.com/2018/03/semantic-image-segmentation-with.html">Semantic Image Segmentation with DeepLab in TensorFlow</a></li> 41 <li><a href="https://medium.com/tensorflow/tensorflow-lite-now-faster-with-mobile-gpus-developer-preview-e15797e6dee7">TensorFlow Lite Now Faster with Mobile GPUs (Developer Preview)</a></li> 42 <li><a href="https://github.com/tensorflow/models/tree/master/research/deeplab">DeepLab: Deep Labelling for Semantic Image Segmentation</a></li> 43</ul> 44
