Texel-Att: Representing and Classifying Element-based Textures by Attributes

Element-based textures are a kind of texture formed by nameable elements, thetexels [1], distributed according to specific statistical distributions; it isof primary importance in many sectors, namely textile, fashion and interiordesign industry. State-of-theart texture descriptors fail to properlycharacterize element-based texture, so we present Texel-Att to fill this gap.Texel-Att is the first fine-grained, attribute-based representation andclassification framework for element-based textures. It first individuatestexels, characterizing them with individual attributes; subsequently, texelsare grouped and characterized through layout attributes, which give theTexel-Att representation. Texels are detected by a Mask-RCNN, trained on abrand-new element-based texture dataset, ElBa, containing 30K texture imageswith 3M fully-annotated texels. Examples of individual and layout attributesare exhibited to give a glimpse on the level of achievable graininess. In theexperiments, we present detection results to show that texels can be preciselyindividuated, even on textures "in the wild"; to this sake, we individuate theelement-based classes of the Describable Texture Dataset (DTD), where almost900K texels have been manually annotated, leading to the Element-based DTD(E-DTD). Subsequently, classification and ranking results demonstrate theexpressivity of Texel-Att on ElBa and E-DTD, overcoming the alternativefeatures and relative attributes, doubling the best performance in some cases;finally, we report interactive search results on ElBa and E-DTD: with Texel-Atton the E-DTD dataset we are able to individuate within 10 iterations thedesired texture in the 90% of cases, against the 71% obtained with acombination of the finest existing attributes so far. Dataset and code isavailable at https://github.com/godimarcovr/Texel-Att