We recently hosted a live webinar to introduce the new DL Workbench and DevCloud integration! See the replay. We received so many great questions live, and I'd like to share our responses to those questions.
- What are other types of use cases supported? Does DL Workbench support other use cases, such as instance segmentation or semantic segmentation? Yes. Use cases supported by OpenVINO are supported by the DL Workbench. Since it integrates the Model Downloader, you can choose from a variety of different models.
- How much file storage does each user get? Is it free? Each user gets 50GB of complementary storage.
- How do I incorporate the performance tuned model generated by the DL workbench into my AI application? With the current version of the application, you can save the model to your local filesystem then upload it to your DevCLoud partition. You can also export a package containing the model and inference engine to build into your application framework.
- What are the available calibration methods? Is it a general-purpose improvement? Do I always need a dataset to calibrate a model? Can we tune calibration algorithms compared to just selecting them in UI? There are various calibration methods, documented here. You don't always need a dataset to calibrate a model - you can use just a few images. Yes, you can tune by just selecting them in the UI.
- How can I ensure security and encrypt my model? Here's a guide
- Can I control which workloads can be shared between VPU and CPU or CPU and FPGA for example? Heterogenous workloads are not currently supported in DL Workbench. However, you can allocate workloads across processors with Jupyter in DevCloud.
- What are the dataset formats supported? Do I have any requirements for a dataset to calibrate on? Should it be different from the validation dataset? Here's documentation about datasets supported.
- Is it availible to mesure throughput on several different CPUs with multidevice with WorkBench? BTW presentation is awesome! Yes, you can test on all the available hardware in DevCloud and compare throughput across devices.
We thank you again for all of your awesome questions that keep us on our toes! We also answered these questions via a live demo, so please take a look at the webinar replay to view the exact step by step.