E'lon
About The Competition
• Competition launch, registration, and proposal submission—runs August 12- 2026; proposals are accepted starting August 13.
• 50 teams will be selected to receive a cloud compute grant valued at $150. All other teams, including those that submit the cloud compute grant proposal remain eligible to continue to the two-month build phase.
Apply for your grant here.
• Final projects are due October 26, 2026, at 11:59 p.m. Pacific Time.
• Every entry must use OpenCV 5 for substantive image or video analysis and run a meaningful component on AWS.
• Two optional featured paths offer separate $1,000 awards: the Best Use of COOL Award and the Agentic Vision Award.
Sponsor: Amazon Web Services (“Amazon” or “Sponsor”)
Administrator: OpenCV Foundation (OpenCV or Administrator, and together with Amazon, the “organizers”)
Join, Build, and Win
Turn an ambitious idea into a working vision application, demonstrate measurable value, and share it with the global OpenCV community.
The competition emphasizes Physical AI and Generative AI systems in which visual understanding drives useful decisions, actions, predictions, or interactions. Entries may pursue either featured path below, combine both, or take another approach and remain fully eligible to win a prize..
Projects may use any programming language or supporting hardware, provided they meet the core OpenCV 5 and AWS requirements on this page.
Build Phase — Two Months
From August 26 through October 26, teams will build
Talablar va baholash
Requirements
Suggested Project Areas
Projects are not limited to these areas, but the organizers especially welcome:
• Active perception for autonomous inspection with agentic orchestration or MCP.
• Physics-informed video prediction for process control.
• Multi-agent visual SLAM with distributed context.
• Real-time spatial digital twins.
• COOL-based server, serverless, container, or hybrid x86/Arm vision pipelines.
• Developer agents that integrate COOL with tools such as Codex, Claude Code, Kiro, or MCP-compatible systems.
• Healthcare, safety, accessibility, agriculture, environmental monitoring, smart cities, education, retail, and sports analytics.
Final Submission Requirements
• A technical report describing the problem, users, architecture, OpenCV 5 implementation, AWS deployment, evaluation, limitations, and responsible-use considerations.
• A public or private judge-accessible code repository or archive. The code does not have to be open source.
• Pinned dependencies plus clear build, deployment, and test instructions.
• An architecture diagram showing the OpenCV 5 and AWS components and, where relevant, COOL or agent components.
• A working web endpoint or an arranged live screen-share demonstration.
• A public or unlisted judge-accessible video of no more than five minutes that shows the team, the application working, its architecture, and its principal results.
• Evaluation evidence appropriate to the project, including failure cases or limitations.
Focus Paths (Special Prizes)
Cloud-Optimized Vision with COOL
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. For Best Use of COOL Award eligibility, COOL must execute the claimed core workload on AWS Graviton or the Arm component of a documented hybrid architecture. Strong submissions will document a reproducible deployment and comparison against an appropriate baseline. Strong submissions will:
• Show that COOL executes the core image or video workload on the Arm path.
• Report reproducible measurements such as latency, throughput, utilization, cost, or developer productivity against an appropriate baseline.
• Explain any x86 and Arm coexistence, container, server, or serverless architecture.
Agentic Vision with OpenCV 5
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. A chatbot that only explains a fixed vision result is not enough—the visual evidence must change what the system does next.
• Examples include autonomous inspection, visual troubleshooting, embodied assistants, active perception, safety monitoring, and human-in-the-loop operations.
• The agent may use any framework or model and may invoke OpenCV 5 and, where applicable, COOL through API
Judging Criteria
Technical execution
weighted 30%: correctness and depth of the OpenCV 5
Innovation
weighted 20%: originality and thoughtful use of computer vision or AI.
Real-world impact
weighted 20%: importance of the problem, usefulness, and evidence of potential benefit.
User experience
weighted 10%: usability, accessibility, and quality of interaction.
Documentation and presentation
weighted 10%: clarity of the report, code, architecture, instructions, and video.
Cloud delivery, reproducibility, and responsible operation
weighted 10%: AWS deployment quality, repeatability, observability, security, and responsible-use practices.