CV Brief · Monday, 15 June 2026
CV Brief
Research & Papers
Morphology-Aware Sample Assignment Fixes IoU Blindspots in Defect Detection
The paper identifies a critical non-sensitive region in IoU scoring where geometrically distinct overlaps produce identical scores, degrading detector training. A morphology-aware assignment strategy corrects this, improving detection quality on surface defect datasets—directly applicable to quality control and manufacturing pipelines.
Read more →Temporal Slot Activation Keeps Inactive Objects Out of Video Decomposition
Proposes TSA to fix unconditional slot propagation in video object-centric learning—slots now activate only when objects are visible, improving temporal consistency. Directly addresses the core inefficiency in video understanding pipelines that process static backgrounds frame-by-frame.
Read more →Pairwise Filter Connections Boost CNN Accuracy Without Architectural Overhaul
Introduces pairwise inter-filter connections as an alternative to pointwise activations, improving CNN performance with minimal architectural change. Practical for practitioners optimizing existing models without retraining from scratch.
Read more →Tutorials & Guides
Why CNNs outperform transformers on image tasks
Technical breakdown of translation invariance and why convolutional architectures maintain advantages for image processing over attention-based models. Essential understanding for practitioners choosing between CNN and transformer-based backbones.
Read more →Getting Started in CV/ML
Deploy YOLOv8 on Raspberry Pi 5 with Hailo acceleration
Step-by-step guide to convert trained YOLOv8/YOLO11 models to HEF format and run inference on Raspberry Pi 5 with Hailo AI HAT+. Practical for edge deployment scenarios where you need full-speed object detection on constrained hardware.
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.