The OpenCV AI Competition 2026 has opened for entries. Competitors will build vision systems that see, reason, and act with OpenCV 5 and Amazon Web Services (AWS).

OpenCV 5 is the latest version of the OpenCV open-source computer vision library. The latest release arrived in June 2026, transforming OpenCV from a simple image-preprocessing helper into a powerful, self-contained AI inference runtime. The latest version eliminates the need to rely on heavy external frameworks like PyTorch or the ONNX Runtime for modern AI models

The competition, which is sponsored by Amazon Web Services (AWS), is a global hackathon in which developers build practical computer vision systems that see, reason, and act using OpenCV 5 and AWS. The competition emphasizes Physical AI and Generative AI systems in which visual understanding drives useful decisions, actions, predictions, or interactions. Any programming language or supporting hardware is allowed. There are $12,000 in main prizes, as well as 50 AWS Compute Grants each valued at $150.

To be in the running for one of the 50 AWS Compute grants, all teams have to do is to write a proposal. The grants are reviewed and awarded on a rolling basis, beginning August 18. To be in the running for one of the AWS Compute grants, entrants are asked to submit a clear problem statement and the real-world impact you intend, along with the OpenCV 5 image or video analysis you plan to build, your planned AWS architecture and services, and a high-level architecture diagram or technical description.

The rules for what the project should do are good. The organizers say that they will especially welcome entries that ise active perception for autonomous inspection with agentic orchestration or MCP. Physics-informed video prediction for process control, or multi-agent visual SLAM with distributed context are also on the especially welcome list, as are real-time spatial digital twins. Entries that use the COOL-based server, serverless, container, or hybrid x86/Arm vision pipelines, and developer agents that integrate COOL with tools such as Codex, Claude Code, Kiro, or MCP-compatible systems will also be favorably considered. The overall areas of interest are healthcare, safety, accessibility, agriculture, environmental monitoring, smart cities, education, retail, and sports analytics.

Programmer'sBookshelf Access

However, competitors can choose to pursue the COOL path, the Agentic Vision path, both, or something else entirely.

Entrants for the Cloud-Optimized Vision with COOL track should use the Cloud-Optimized OpenCV Library (COOL), available through AWS Marketplace and optimized for AWS Graviton, to accelerate widely used vision operations and move from prototype to scalable cloud deployment. Strong submissions will document a reproducible deployment and comparison against an appropriate baseline.

Alternatively, entrants to the Agentic Vision with OpenCV 5 track should build a workflow in which an agent uses OpenCV 5 tools within a multi-step perception-decision-action loop. To qualify, image or video results must influence a subsequent plan, tool call, action, or request for human approval. The rules say that a chatbot that only explains a fixed vision result is not enough - the visual evidence must change what the system does next.

The build phase runs from August to October, and covers the build and evaluation stages, as well as documentation and submitting the project. The first prize winner will receive $5,000, with a second prize of $3,000, a third prize of $2,000, and two $1,000 special awards.