Aberrations in optics for CV: analysis of defects https://sudonull.com/aberrations-in-optics-for-cv-analysis-of-defects
Let's break down spherical aberration, coma, astigmatism and other effects in machine vision. How they affect sharpness and how to correct for computer vision tasks. For developers.
Robot CUE7 Toyota: Breakthrough in Physical AI https://sudonull.com/robot-cue7-toyota-breakthrough-in-physical-ai
Toyota tests machine vision and control through basketball robot CUE7. Learn about weight reduction, records, and applications in robotics. Read the analysis of implications for the industry.
TAPe detection: 100k parameters vs SOTA models https://sudonull.com/tape-detection-100k-parameters-vs-sota-models
How the TAPe detector with 100k parameters outperforms YOLO and RF-DETR in speed and resource savings. Detailed breakdown of benchmarks and advantages for developers.
TAPe on COCO: 98% accuracy with boxes https://sudonull.com/tape-on-coco-98-accuracy-with-boxes
TAPe achieved 98% on 2% COCO, bounding boxes instead of centroids, minimum false positives. Experiment details for CV developers. Study the approach.
TAPe embeddings: 74% on COCO without transformers https://sudonull.com/tape-embeddings-74-on-coco-without-transformers
Training TAPe embeddings on synthetics gives 74% COCO classification. Rejection of transformers, reconstruction results 82%. For CV developers — experiment details and plans.