CV Brief · Monday, 2 November 2026
CV Brief
Research & Papers
PathoJev: Batch pathology decisions from shared visual context
PathoJev predicts answer probabilities for multiple pathology assessments (tumor type, grade, invasion) directly from whole-slide images without generating separate textual responses for each question. Reduces latency between visual evidence and structured decisions—critical for clinical pathology pipelines handling high-throughput slide analysis.
Read more →AquaProbe: Adapt frozen vision models for underwater scenes efficiently
AquaProbe adapts pretrained vision models to underwater imagery by selectively applying adapters to encoder layers based on domain divergence, handling turbidity, backscatter, and wavelength attenuation. Directly applicable to autonomous underwater robotics and marine survey pipelines without full model retraining.
Read more →Sparse autoencoders interpret histopathology foundation model embeddings
Uses sparse autoencoders to decompose dense embeddings from pathology foundation models into interpretable morphological concepts—tissue architecture, cellular morphology, staining artifacts. Enables practitioners to validate and debug model decisions at concept level for clinical deployment.
Read more →Tutorials & Guides
Serving ResNet-50 on Kubernetes with Triton: Dynamic Batching
Dynamic batching in Triton inference server significantly improves GPU utilization when serving ResNet-50 models on Kubernetes. Demonstrates practical production deployment patterns for maximizing throughput on expensive GPU infrastructure.
Read more →YOLO Tracking: ByteTrack vs BoT-SORT Parameter Tuning
Hands-on comparison of ByteTrack and BoT-SORT trackers within Ultralytics YOLO, with parameter explanations and real PPE detection experiments. Essential reference for practitioners choosing and tuning multi-object tracking in production pipelines.
Read more →When extracting crops from CCTV at scale, always use frame seeking (cv2.CAP_PROP_POS_FRAMES) instead of sequential reads. On a 2-hour video at 1FPS you'll go from hours to minutes.
Quick Links
- DyCraft: An Agentic System for Long-Horizon 4D Scene Generation
- Drone-PUR: From Visual Clue Discovery to Understanding and Reasoning in UAV Scen
- EgoHSI3R: Online 4D Hand-Scene Interaction Reconstruction from Egocentric Video
- Spatial Atlas-Informed Loss Ensembles for Whole-Body PSMA PET/CT Metastasis Segm