ISSUE #004
DRAFTING
Computer Vision & Image Processing
Silicon Sight
A 3D eye with bounding boxes for eyelashes. We taught a sphere to see, and now it judges your code.
WHAT’S BREWING
- Why end-to-end ViTs are eating CNN-era pipelines, and what they break in the process.
- Bounding boxes are dead, segmentations are dying, the future is open-vocabulary scene understanding.
- DINOv3, SAM-3, EVA-02: the self-supervised line that makes labels feel like a 2019 affectation.
- When your model has to look at a screen as well as the pixels: GUI-aware vision and the rise of agent-vision benchmarks.
- What an OS-level perception primitive looks like: 'recall every image of my child wearing a yellow hat' as a single syscall.