Why does it take so long to develop new computer vision algorithms?
Engineering tasks like model training, running inference, performance analysis, and integration into AI pipelines are often conducted in separate places, on local machines or various platforms in the cloud. This makes collaboration and asset sharing for computer vision teams unnecessarily time-consuming and difficult.
The Rendered.ai Platform as a Service already provides collaborative workspaces for rapid synthetic data generation with configurable and reusable automated workflows and streamlined integration into AI pipelines. The 1.4.0 Rendered.ai Platform update offers the following features to accelerate computer vision development and boost effective team collaboration.
Machine Learning
Rendered.ai users can now access and train computer vision models with their own datasets and/or synthetic datasets generated on the platform. Best-in-class models for applications like classification and object detection are available. Users can also work with the Rendered.ai team to integrate other preferred computer vision models.
Configuring model training parameters and monitoring the progress of each training job run on the platform is easy.
A new training menu to easily configure model hyperparameters in the Rendered.ai Platform as a Service.
Rendered.ai Platform dashboard for monitoring model training jobs in real time.
To see an example of how to use this new feature, check out our tutorial on training a Detectron2 computer vision model on the Rendered.ai Platform.
Inference Services
Another newly added feature for amplifying model training in the Rendered.ai Platform is running inference to make predictions against synthetic datasets generated or real datasets uploaded to the platform.
After an inference job is run on the Rendered.ai Platform, the results can easily be viewed in an inference library. If ground truth annotations were provided with the dataset, then the following inference metrics will also appear:
- Overall inference metrics
- Class-specific metrics for Precision, Recall, F1 and mAP
- Plots for a Confusion Matrix, PR Curve, and ROC Curve
Rendered.ai Platform inference library with detailed analytics.
A dataset image viewer inside the platform also enables you to view the predictions from the inference job run with the ability to toggle the ground truth, inference layers, and prediction confidence labels on and off as desired.
A newly added dataset image viewer to visualize different layers of inference.
Ready to see these new features in action? Sign up for a free trial or login to the Rendered.ai Platform with content code: MARINECLASS to access a workspace pre-loaded with assets and generate customized synthetic data to train a NVIDIA TAO classification model.
