|
|
---
|
|
|
license: cc-by-4.0
|
|
|
---
|
|
|
|
|
|
# Introduction to Anywhere3D-Bench(anywhere3d_v2)
|
|
|
|
|
|
anywhere3d_v2.json contains the all 2886 referring expressions-3D bounding box pairs from Anywhere3D benchmark. It's based on the first version of [Anywhere3D dataset](https://huggingface.co/datasets/txwang98/Anywhere3D), which contains 2632 referring expressions-3D bounding box pairs. We increased the number of annotations for the two most challenging grounding levels in Anywhere3D-Bench: the space level and the part level. Specifically, 106 annotations were added to the space level and 148 to the part level, with a particular focus on expanding the most difficult movement tasks within the part level.
|
|
|
|
|
|
The meaning of each key in the dictionary is as follows. The first table contains the fields that directly related to expressions-3D bounding box pairs while the second table contains fields that are helpful for annotation interface.
|
|
|
|
|
|
|
|
|
|
|
|
| Field Name | Description |
|
|
|
|------------------------|-----------------------------------------------------------------------------|
|
|
|
| grounding_level | Grounding level of the expression, selected from one of the four values: ["area_level", "space_level", "object_level", "part_level"]. |
|
|
|
| expression_type | Specifies the type of expression at each grounding level. |
|
|
|
| datasetname | Selected from one of the four values: ["scannet", "multiscan", "3RScan", "arkitscene_valid"]. |
|
|
|
| scene_id | From simplicity during Human Annotation and Verification, we rename the scene_id in the original dataset. |
|
|
|
| scene_name | The original scene_id in four datasets. |
|
|
|
| referring_expressions | Referring expressions in Anywhere3D. |
|
|
|
| box_width | The length of the ground-truth 3D bounding box along X-axis. |
|
|
|
| box_length | The length of the ground-truth 3D bounding box along Y-axis. |
|
|
|
| box_height | The length of the ground-truth 3D bounding box along Z-axis. |
|
|
|
| box_x | The center coordinate of the ground-truth 3D bounding box along X-axis. |
|
|
|
| box_y | The center coordinate of the ground-truth 3D bounding box along Y-axis. |
|
|
|
| box_z | The center coordinate of the ground-truth 3D bounding box along Z-axis. |
|
|
|
| box_rot_angle | The rotation angle of ground-truth 3D bounding box. |
|
|
|
|
|
|
|
|
|
<!-- <span style="color:red">"grounding_level"</span>: grounding level of the expression, selected from one of the four values: ["area_level", "space_level", "object_level", "part_level"].
|
|
|
|
|
|
<span style="color:red">"expression_type"</span>: specifies the type of expression at each grounding level.
|
|
|
|
|
|
"db_id": specifies the database ID in MongoDB during Human Annotation and Verification.
|
|
|
|
|
|
"expressions_cnt_in_annotation_interface": specifies the expression number during Human Annotation and Verification.
|
|
|
|
|
|
<span style="color:red">"datasetname"</span>: selected from one of the four values: ["scannet", "multiscan", "3RScan", "arkitscene_valid"].
|
|
|
|
|
|
<span style="color:red">"scene_id"</span>: From simplicity during Human Annotation and Verification, we rename the scene_id in the original dataset.
|
|
|
|
|
|
<span style="color:red">"scene_name"</span>: The original scene_id of the four datasets.
|
|
|
|
|
|
<span style="color:red">"referring_expressions"</span>: referring expressions in Anywhere3D.
|
|
|
|
|
|
<span style="color:red">"box_width"</span>: The length of the ground-truth 3D bounding box along X-axis.
|
|
|
|
|
|
<span style="color:red">"box_length"</span>: The length of the ground-truth 3D bounding box along Y-axis.
|
|
|
|
|
|
<span style="color:red">"box_height"</span>: The length of the ground-truth 3D bounding box along Z-axis.
|
|
|
|
|
|
<span style="color:red">"box_x"</span>: The center coordinate of the ground-truth 3D bounding box along X-axis.
|
|
|
|
|
|
<span style="color:red">"box_y"</span>: The center coordinate of the ground-truth 3D bounding box along Y-axis.
|
|
|
|
|
|
<span style="color:red">"box_z"</span>: The center coordinate of the ground-truth 3D bounding box along Z-axis.
|
|
|
|
|
|
<span style="color:red">"box_rot_angle"</span>: The rotation angle of ground-truth 3D bounding box. -->
|
|
|
|
|
|
|
|
|
|
|
|
| Field Name | Description |
|
|
|
|-------------------------------------|-----------------------------------------------------------------------------|
|
|
|
| db_id | Specifies the database ID in MongoDB during Human Annotation and Verification |
|
|
|
| expressions_cnt_in_annotation_interface | Specifies the expression number during Human Annotation and Verification |
|
|
|
| cylinder_diameter | The diameter of the distance measurement tool |
|
|
|
| cylinder_length | The length of the distance measurement tool |
|
|
|
| cylinder_x | The center coordinate of the distance measurement tool along X-axis |
|
|
|
| cylinder_y | The center coordinate of the distance measurement tool along Y-axis |
|
|
|
| cylinder_z | The center coordinate of the distance measurement tool along Z-axis |
|
|
|
| cylinder_rot_angle | The rotation angle of distance measurement tool |
|
|
|
| window_camera_position_x | The camera position (x) in annotation interface |
|
|
|
| window_camera_position_y | The camera position (y) in annotation interface |
|
|
|
| window_camera_position_z | The camera position (z) in annotation interface |
|
|
|
| window_camera_quaternion_x | The camera quaternion (x) in annotation interface |
|
|
|
| window_camera_quaternion_y | The camera quaternion (y) in annotation interface |
|
|
|
| window_camera_quaternion_z | The camera quaternion (z) in annotation interface |
|
|
|
| window_camera_quaternion_w | The camera quaternion (w) in annotation interface |
|
|
|
| window_camera_target_x | The camera target (x) in annotation interface |
|
|
|
| window_camera_target_y | The camera target (y) in annotation interface |
|
|
|
| window_camera_target_z | The camera target (z) in annotation interface
|
|
|
|
|
|
|
|
|
<!--
|
|
|
"cylinder_diameter": The diameter of the distance measurement tool.
|
|
|
|
|
|
"cylinder_length": The length of the distance measurement tool.
|
|
|
"cylinder_x": The center coordinate of the distance measurement tool along X-axis.
|
|
|
|
|
|
"cylinder_y": The center coordinate of the distance measurement tool along Y-axis.
|
|
|
|
|
|
"cylinder_z": The center coordinate of the distance measurement tool along Z-axis.
|
|
|
|
|
|
"cylinder_rot_angle": The rotation angle of distance measurement tool.
|
|
|
|
|
|
"window_camera_position_x", "window_camera_position_y", "window_camera_position_z", "window_camera_quaternion_x", "window_camera_quaternion_y", "window_camera_quaternion_z", "window_camera_quaternion_w": The camera position and quaternion in annotation interface.
|
|
|
|
|
|
"window_camera_target_x", "window_camera_target_y", "window_camera_target_z": The camera target in annotation interface. -->
|
|
|
|
