All the release notes for Voyager updates go straight out on the GitHub repo, so I don’t tend to parrot them here, but this is a pretty big one and feels like it deserves a spot here on the community!
If you’ve been using 1.3 already, there are some nice changes that make life easier.
A few things that particularlt stand out to me:
- 🔹 Cascade pipelines feel smoother. Chaining models together (like running a classifier on detections) takes less fiddling around than before.
- 🔹 More operator and metadata options. You can now shape pre/post-processing and inference results in ways that fit better with your own app.
- 🔹 Model Zoo is easier to adapt. Swapping in your own weights on the reference pipelines is less hassle, so you can get something tuned to your dataset faster.
- 🔹 Custom evaluators are handy if you want accuracy measured in a way that matches your own project rather than sticking with defaults.
- 🔹 General quality-of-life polish in pipeline debugging, deployment consistency, benchmarking… all just smoother.
- 🔹 You can never have too many new models. YOLO10[n,s,b], FastSAM, FaceNet, OSNet and Deep-OC-Sort can now be used within Voyager pipelines.
If you’ve been on 1.3, the upgrade feels like a nice step up rather than just “another version bump”.
Full notes are here for the deep dive: Voyager SDK v1.4.0 Release Notes
Curious if anyone else has tried it yet? Which of the new features are you putting to use first?
