Netgraf
Draw who knows whom, then find the shortest chain between any two people
Emerged finds new products worth a look, daily — what’s new today
RelateAnything
Open-vocabulary relation prediction between any image regions, in real time
Robotics and vision teams can ask how objects relate using any predicate phrase at inference, in about 20 ms a frame, without retraining a scene-graph model.
RelateAnything predicts relationships between regions of an image, such as boxes or masks from any detector, with the predicate vocabulary supplied as text at inference. It pairs a DINOv3 backbone with a relation transformer and text encoder, in models from 46M to 114M parameters. The author reports about 20 ms per frame on an A40 and 7 fps on CPU via OpenVINO. Weights and the RA-4M dataset of 4.2 million relations are on Hugging Face, with an arXiv paper and a fully in-browser demo. Code is Apache-2.0; weights follow the DINOv3 licence.
More like this
WeMM-Embedding
Open multimodal embedding models from Tencent's WeChat Vision Team
VisualFries
An MIT render engine your coding agent drives to cut raw footage into posts
InstinctFlash
A serving runtime that speeds robot policy models up for real control loops
glassbrain
A browser viewer and diff tool for the knowledge graphs your agents write
Warp
Run trillion-parameter models on a laptop by streaming weights from NVMe