[Feature Request] Add C++ sample for Live VLM Chat using Webcam (OpenCV)

[Ashitpatel001]

Problem

Currently, the visual_language_chat sample only supports static images loaded from a file path.

While this demonstrates the API, it does not showcase the real-time performance capabilities of OpenVINO GenAI on edge devices. Developers building visual agents or robots need a reference implementation for handling continuous video streams without blocking the inference loop.

Proposed Solution

I propose adding a new C++ sample: live_vlm_chat.
This sample integrates OpenCV to capture a live webcam feed and interacts with the VLM (e.g., LLaVA/Mistral) in real-time.

Key Features of the proposed sample:

• Multi-threaded Architecture: Decouples the UI/Camera loop (Main Thread) from the Inference loop (Worker Thread) to ensure the video feed never freezes while the LLM is "thinking."
• Thread Safety: Implements std::mutex and std::condition_variable to safely pass frames between threads.
• MSVC Compatibility: Solves the C3889 build error on Windows by strictly defining tensor types (std::vector<ov::Tensor>) for the ov::genai::images constructor.
• Interactive UI: Allows users to "snap" a frame and chat with it while the camera continues running.

Implementation Details

I have already implemented and tested this locally on Windows 11 (Intel CPU & iGPU).

• Dependencies: Adds OpenCV (core, highgui, videoio) as an optional dependency in CMakeLists.txt.
• File: samples/cpp/visual_language_chat/live_vlm_chat.cpp

I have a working implementation ready and tested. I can submit a Pull Request immediately if this contribution aligns with the project's roadmap.

[github-actions]

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

[Ashitpatel001]

Hi team,

I have submitted PR #3308 which fully implements this feature request. It adds the live_vlm_chat.cpp sample with OpenCV webcam integration as described above.

The code is tested on Windows (Intel CPU/iGPU) and is ready for review. Could you please take a look? I would love to get this merged to expand the C++ VLM capabilities. Thanks!