Computer Vision News - October 2018

To get more of a hands-on understanding and a better grasp on research papers and methods, you should look under the hood -- at the code itself. One line of code can, quite often, be worth a thousand pages of explanations. We’ll use the DeepLabV3+ network for semantic segmentation and its implementation using Keras, as our example today. We’ll focus on 5 central parts of the code, marked by roman numerals I-V in the figure below, reviewed in the other article this month (see page 4). You can find the (unofficial) Keras implementation here . The most essential code, which our review will focus on, is found in the Deeplabv3 function in the file model.py defined in line 275. The official implementation can be found under model/research of TensorFlow here ; however, the TensorFlow implementation is more generic and harder to understand. 56 Focus on: DeepLabV3+ in Keras Tool by Assaf Spanier “…look under the hood -- at the code itself!” Computer Vision News