Dataset Card for Dataset Name This is a FiftyOne dataset with 27083 samples. Installation If you haven't already, install FiftyOne: Usage Dataset Details Dataset Description The QMUL OpenLogo dataset is a large scale logo detection benchmark designed for evaluating open logo detection algorithms in realistic deployment scenarios. Unlike traditional logo detection datasets that assume exhaustive fine grained labeling for all classes, QMUL OpenLogo simulates real world conditions where new logo classes arrive progressively and require detection with little or no labeled training data. The dataset was created by aggregating and refining 7 existing logo detection datasets to establish a comprehensive benchmark for open logo detection. It contains 27,083 images across 352 logo classes, with a distinctive train/test split that designates only 32 classes as "supervised" (with labeled bounding boxes) while the remaining 320 classes are "unsupervised" (having only clean logo design images). Curated by: Queen Mary University of London (QMUL) and Vision Semantics Limited Funded by: China Scholarship Council, Vision Semantics Ltd, Royal Society Newton Advanced Fellowship Programme (NA150459),…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy