README.md

---
dataset_info:
  features:
  - name: object_name
    dtype: string
  - name: object_room
    dtype: string
  - name: object_id
    dtype: int32
  - name: num_images
    dtype: int32
  - name: views
    sequence: int32
  - name: images
    sequence: image
  - name: image_paths
    sequence: string
  - name: depth_maps
    sequence: image
  - name: depth_paths
    sequence: string
  - name: segmentation_images
    sequence: image
  - name: segmentation_matrices_all
    sequence:
      array2_d:
        shape:
        - 512
        - 512
        dtype: int32
  - name: segmentation_matrices_single
    sequence:
      array2_d:
        shape:
        - 512
        - 512
        dtype: int32
  - name: segmentation_paths
    sequence: string
  - name: segmentation_matrix_all_paths
    sequence: string
  - name: segmentation_matrix_single_paths
    sequence: string
  - name: camera_positions
    sequence:
      array2_d:
        shape:
        - 1
        - 3
        dtype: float32
  - name: camera_poses
    sequence:
      array2_d:
        shape:
        - 4
        - 4
        dtype: float32
  - name: bbox2d_projected
    sequence:
      array2_d:
        shape:
        - 2
        - 2
        dtype: int32
  - name: bbox3d_projected
    sequence:
      array2_d:
        shape:
        - 8
        - 2
        dtype: float32
  - name: bbox2d_tight_visible_part
    sequence:
      array2_d:
        shape:
        - 2
        - 2
        dtype: int32
  - name: bbox2d_tight_whole_object
    sequence:
      array2_d:
        shape:
        - 2
        - 2
        dtype: int32
  - name: visibles
    sequence: bool
  - name: in_views
    sequence: bool
  - name: visible_ratios
    sequence: float32
  - name: visible_pixels
    sequence: int32
  splits:
  - name: train
    num_bytes: 17141032954
    num_examples: 604
  download_size: 1823673031
  dataset_size: 17141032954
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
task_categories:
- image-to-text
tags:
- VLM
- 2D
size_categories:
- 1B<n<10B
---

Dataset used for 3D understanding in VLMs from multiview 2D images.