We can provide all teams with free GPU access and other cloud based tools for this competition. You can build / experiment / train / and deploy serverless APIs - so that your edge computing hardware can continually get the latest pickled models. Imagine being able to just collect video non stop in the daytime, annotators label video at night, model gets trained after QC, you wake up and the retrained model is pushed to the drone before you even wake up to collect more video data the next day.
GPU Optimized Annotation Workflow:
GPU optimized annotation and labeling for video and images. Cost effectively use remote low cost annotators (some start at $2.50 an hour on upwork).
Custom Annotation Models:
You can customize your bounding box tensorflow annotation models (i.e., imagenet, openimages, etc...)
Parallel Model training on GPU clusters while easily parameter sweeping:
You can spin up multiple parallel jobs to train models with a click of a button. Train your long running models in parallel without constantly setting up new docker environments. You can easily perform hyper parameter sweeping across parallel jobs by passing arguments into each job.
Here's an example:
Git Version Control Code:
All of your code is version controlled using local or git or github. Simply push up your code and all workspaces and jobs will have access.
Team Based Collaboration:
You can easily add new members to teams and make workspaces public or private with the click of a button - no more AWS keys to share.