{
  "id": "2204.08479",
  "version": "v1",
  "published": "2022-04-18T17:34:37.000Z",
  "updated": "2022-04-18T17:34:37.000Z",
  "title": "Inductive Biases for Object-Centric Representations of Complex Textures",
  "authors": [
    "Samuele Papa",
    "Ole Winther",
    "Andrea Dittadi"
  ],
  "categories": [
    "cs.CV",
    "cs.LG"
  ],
  "abstract": "Understanding which inductive biases could be useful for the unsupervised learning of object-centric representations of natural scenes is challenging. Here, we use neural style transfer to generate datasets where objects have complex textures while still retaining ground-truth annotations. We find that, when a model effectively balances the importance of shape and appearance in the training objective, it can achieve better separation of the objects and learn more useful object representations.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2022-04-18T17:34:37.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "complex textures",
      "object-centric representations",
      "inductive biases",
      "neural style transfer",
      "achieve better separation"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}