Computer Vision News - January 2019

For each run of your code, comet.ml will report run outcomes, results and the values of whichever parameters you define. You list the parameters you want comet.ml to monitor using the log_parameter function. For example, in the snippet below we instruct comet.ml to monitor batch_size : Comet.ml allows us to monitor and analyze training and testing separately, using the functions Experiment.train() and Experiment.test() , which define separate contexts for training and testing; there is also an equivalent function for validation -- Experiment.validate() . Now, we’ll look at a complete example, using Keras (in the code below we omit the import and data loading -- to save space). The code starts by defining the Experiment class , which connects to the comet.ml website. We then define 6 parameters for monitoring (batch size, epochs, etc). Next, we define a really basic two-layer network. And finally we run training and then testing sessions using the appropriate Experiment.‘session’() functions. 21 Focus on Computer Vision News Tip - Train Your Network from comet_ml import Experiment #if "my project name" does not already exist, it will be created. experiment = Experiment(api_key="YOUR_API_KEY", project_name="my project name") batch_size = 4 experiment.log_parameter("batch_size", batch_size) Feedback of the Month Just before releasing out first version of the product, we encountered new data, and needed to rapidly develop an algorithm for head CT basic segmentation. RSIP Vision were very professional, practical and responsive . They provided a solution quickly and improved it to work on new data we sent them. The communication before and during the project was excellent . The algorithm is still used today in our product , doing a good job on hundreds of CT scans. Yoav Pinsky Software Engineer - Biosense Webster, part of Johnson & Johnson