Voyager v1.7 landed and, past the usual new-models-and-hardware stuff, there are a few things that change what you can actually build on Metis that I thought it was worth us all taking two minutes to look at.
The one that caught my eye right away is that re-identification is now a first-class task type. ReID is what lets you say "that's the same person who walked past the front camera two minutes ago," rather than treating every camera and every frame as a fresh unknown. If you've built a Home Assistant camera notifier and ended up hand-rolling logic to (try and) stop it pinging you about the same person fifteen times (which is something I’ve been battling with a lot personally), this is the release that lets you bin that code. There's a new tracking_with_classification.py demo that chains detection to tracking to per-track classification, which is basically the skeleton of a solid smart-security pipeline. Very cool.
Depth estimation from a single camera is in there too, with a FastDepth demo. Not using a stereo rig or LiDAR, that is. I mean from one ordinary camera and you can get a depth map. Handy for obstacle detection and basic spatial awareness on edge robotics. It’s not going to replace stereoscopic vision or active depth perception, but if you just need to push data into the third dimension a little bit, this is awesome. And oriented bounding boxes (OBB) mean you can finally detect things at an angle without a normal box swallowing half the background: aerial footage, shelf items, text on packaging.
Under all of that, the Python-native Pipeline Builder grew up. There's now a proper scheduler that spreads work across the AIPU cores, pipeline.stream to pipeline frames automatically, and a built-in zero-copy video decoder (cv.create_source). So you can build a real multi-stage streaming app in plain Python, no GStreamer YAML wrangling (unless you want it, of course). There's even a single-core debug mode that runs each frame end-to-end in the calling thread so you can step through your operators instead of guessing.
Model-wise, we’ve looking at YOLO26 semantic segmentation (n through x), YOLOv4 bringing Darknet format support, and MobileNetV3. All via axdownloadmodel as usual.
If you build something with the new ReID or depth support, post it. I'd really love to see what people do with cross-camera tracking especially!