ReTexture 2D

The 2D outputs seen here are silhouettes of the 3D models from ReTexture 3D dataset (10.34740/kaggle/dsv/9877655). The 3D models employed in this dataset to create the 2D silhouettes were sourced from ShapeNet. We utilized ten categories of models—five biological and five non-biological. The biological categories were fish, elephant, butterfly, bird, and bear, while the non-biological categories were helmet, mailbox, bathtub, mug, and phone. For each of the ten selected categories, we selected 10 high-quality 3D models and paired them with ten representative textures specific to each category. We automated the stimuli creation process using BlenderAPI. We imported models from ShapeNet and consolidated each into a single, unified mesh. The object was then unwrapped onto its corresponding texture, after which the texture was reapplied to the model. All of the 100 models were wrapped in all of the 100 textures, as well as a solid black texture; wrapping in the solid black texture generated the silhouettes necessary to produce the ReTexture 2D dataset. All 100 textures were flatly mapped onto these silhouettes using binary masks and background subtraction. This process generated 120,000 total 2D images (the 2D ReTexture dataset).