Custom-Built Computer Vision: Robovision Software Development Kit
A Closer Look at the Robovision SDK
In computer vision, where use cases vary widely and requirements shift with each production line, off-the-shelf tools can only go so far. The Robovision Software Development Kit (SDK) was built to meet the demands of teams who need more control—those who have proprietary models, edge-case data, or infrastructure constraints that fall outside generic frameworks.
The SDK extends the core Robovision Platform—originally designed to guide users from raw image data through labeling, training, testing, and live inference—by giving data scientists the tools to adapt each stage to local realities.
Built for Integration and Adaptation
The SDK and accompanying REST API allow full integration of custom algorithms, proprietary data sources, and internal systems. Whether you need to interface with a pre-existing metadata store, configure imports from a cloud-based data lake, or write custom logic for domain-specific tasks, the SDK accommodates it.
Workflows can be modified or extended at any stage. Custom preprocessing pipelines, evaluation metrics, and monitoring tools can be layered directly into the platform. Teams are not confined to a preset mold—they can add to it, restructure it, or replace parts of it entirely.
Beyond the Default: Custom Blocks and Algorithm Extensions
Robovision’s architecture is organized around four core modules—Label, Train, Test, and Inference. With the SDK, teams can insert custom modules between or alongside these default steps. These additions, referred to as algorithm extensions, support both pre- and post-processing logic. There is no hard limit on the number or type of blocks that can be added.
You might add a metadata parser that maps values from a third-party labeling tool. Or configure a wizard to fine-tune inference parameters. Or introduce domain-specific optimization logic that automatically tunes hyperparameters based on sensor input. All of this is possible within the SDK framework.
Once configured, custom flows can be packaged, named, and surfaced directly in the Robovision interface for easy deployment and handoff—whether internally or to external partners.
Modular Workflow Extensions
Add Custom Blocks to the AI Lifecycle
Go beyond the default Label–Train–Test–Infer structure. Insert pre- and post-processing steps, metadata importers, or domain-specific logic as standalone modules. No imposed limits on the number of custom blocks.
Deep Integration with Existing Infrastructure
Connect to Your Data and Systems Directly
Use the SDK and REST API to link proprietary data sources, cloud storage (e.g., Azure data lakes), or internal tools. Configure imports, triggers, and logic tailored to your operational setup.
Refine and Improve
Retraining is simple, with one-click auto-generated masks using foundation model predictions. Use custom dashboards to track performance and retrain at the optimal time for maximum accuracy.
Adapt and Extend: A Toolkit for Custom Computer Vision Workflows
Technical Autonomy, Not Platform Lock-In
The Robovision SDK is intended for organizations that take AI seriously—teams with long-term investments in internal models, subject-matter constraints, or hardware-specific tuning requirements. It provides the autonomy to shape AI pipelines around existing operational workflows, rather than the other way around.
Designed for Scientists, Not Just Users
Not a black box., the SDK is Python-based and built for those who prefer to work close to the code. It does not dictate how your algorithm should function, nor does it assume your infrastructure follows a common path. Whether you are building on top of Robovision’s models or deploying your own from scratch, the platform does not stand in your way.
If the default pipeline works for your use case, you can use it as is. If it does not, you can reconfigure it entirely.
Robovision in Action: Success Stories
System integration partner QING used Robovision's AI platform to automate its customer's quality control.
View success story
Computer Vision AI integrated into a tulip planter machine
View success story
See how the robot trims rose cuttings from a branch and pots them in one operation.
View success story
PackCheck integrates AI platform to improve and scale quality inspection in food and packaging
View success story