We look at the discovery of 3D internal item in relation to one RGB (-D) framework found on the Commodity Handheld device. We want to progressively to the status of the data and modeling. First, we develop that existing datassets with high limit in moderation, accuracy, and variation. As a result, we present cubify-after 1m (CA-1M), making fun of 400k 3D items in 1k reliable trust-related scenes with complete 3.5k captivity, EGOFTIC Capture. Next, inventing cubify transformer, the basis of the acquisition of transformer 3D substance to the 3D object in the Point or Voxel. (-D) Input from RGB (-D) Input from RGB (-D) input. While this method does not have a 3D discrimination, it shows that the can-1m, recall 62% of the 3D depth maps, and they provide only the deeper. In addition, with previous training in CA-1M, cutting methods out of the unique POINT separately the Sun RGB-D – Supporting the Small 3D size is useful for the bad CA-1M information. Overall, this data and model model and model's basis provides strong evidence that we move forward to models that can well do anything.
