PeekingDuck

PeekingDuck packages input, model, post-processing, and output into nodes. Developers can write a YAML flow such as "camera input -> YOLO detection -> draw boxes -> screen output" without building from raw PyTorch / TensorFlow primitives.

It fits teaching, rapid prototyping, and SME CV applications such as footfall counting, object detection, pose analysis, and safety-compliance checks.

PeekingDuck’s point is not chasing the newest model, but lowering the deployment threshold for computer vision. Many organizations need a pipeline that can run, be adjusted, and ship, rather than starting from paper reproduction.

That product logic matches AISG’s applied orientation: package AI capability into tools engineers can actually use.

PeekingDuck is part of Singapore’s open-source AI tooling line. Like TagUI, it serves the goal of making AI easier for local enterprises to adopt.

Its long-term tracking point is community activity and positioning: as multimodal models advance quickly, classical CV pipeline frameworks must either move toward edge and industrial deployment, or risk being absorbed by more general multimodal tools.