
CDIKTNet Localizes Drones From Satellite Imagery With Just 2% Labeled Data
A new drone-view geo-localization network learns from as little as 2% paired data and transfers to new environments without any pairing labels.
20 stories

A new drone-view geo-localization network learns from as little as 2% paired data and transfers to new environments without any pairing labels.

A two-stage training recipe teaches vision-language-action driving models to reason step by step and self-correct with physics-grounded rewards.

A lightweight pipeline pairs a learned frame selector with an attention-boosted YOLOv8, cutting compute up to 23x while keeping wildfire detection accurate.

A new benchmark of 42K fetal ultrasound images and 93K questions reveals the best vision-language model manages just 55% accuracy.

A KAIST team turns optical flow into a geometry supervisor for monocular 3D Gaussian Splatting SLAM, beating depth-prior methods on tracking accuracy.

A new 170K-image benchmark and a lightweight noise-injection baseline take aim at applying open-vocabulary segmentation to satellite and drone imagery.

A new framework brings causal inference to unsupervised cross-modality person re-identification, closing the labeling gap for 24/7 surveillance video.

A new 41,307-pair benchmark forces vision-language models to localize findings across MRI slices, exposing a large gap between describing anatomy and pointing to it.

A training-free calibration method fuses cosine and Mahalanobis distances on frozen CLIP embeddings to fix prototype drift across imbalanced, heterogeneous visual domains.

A training-free trick reshapes the causal mask at inference time, forcing vision-language models to ignore distracting image regions and answer more accurately.

A new paper proposes fuzzy-logic reward shaping for reinforcement learning, cutting training epochs and variance on hard drone racing tracks.

A search-based planner distinguishes helpful contact from dangerous contact, letting a two-segment catheter autonomously navigate three aortic arch types with 100% success.

Researchers publish an agent-based simulation platform that embeds metaheuristic optimizers directly into the loop, trimming search cost by 82% for charger and solar planning.

A gesture-driven annotation tool replaces click-heavy labeling workflows with hand signs, hitting 0.923 F1 and winning over 88.9% of testers.
Black Forest Labs shipped a distilled VAE decoder for FLUX.2 that runs 1.4x faster and cuts peak VRAM by roughly the same margin.

OBLITERATUS shipped a fully uncensored fine-tune of Google's new Gemma 4 E4B, driving hard refusals from 98.8 percent to zero while keeping the model runnable on a phone.

Alibaba's new open-weight MoE activates only 3B of 35B parameters per token yet hits 73.4% on SWE-bench Verified.

A single image goes in, a decomposable multi-part 3D mesh comes out. No segmentation, no two-stage pipeline, invisible parts included.

Snyk's open-source agent-scan CLI audits your local MCP servers and AI agent skills for prompt injection, tool poisoning, and hardcoded secrets before they can be exploited.
Portable-AI-USB bundles Ollama, Llama 3, and AnythingLLM onto a flash drive for a fully offline, zero-trace AI that leaves no footprint on any host machine.