maximiliangilles
commented
1 day ago Hi @milenayahya,
Thank you for your question! For detailed guidance on how to read in the ground truth, I recommend checking the source code of ./Scripts/visualize_2d.py. It contains all the relevant information you'll need.
Please note that the 2D bounding boxes are not explicitly included in the dataset. However, they can be derived from the pixelwise instance segmentation masks provided.
Let me know if you have any further questions!
Best regards, Max
Hello, thank you for your meticulous work. I need the data annotations, in particular I need the bounding boxes and categories. I tried to read the ".../scene/_scene.hdf5" in 3 different ways, and the content of these files is always: camera camera/pose_relative_to_world objects objects/categories
As opposed to the expected: |>camera |>poses_relative_to_world |>keypts |>com |>keypts_relative_to_camera |>object_id |>byhand |>keypts_relative_to_camera |>object_id |>occluded_keypts_relative_to_camera |>occluded_object_id |>non_colliding_grasps |>paralleljaw |>contact_poses_relative_to_camera |>contact_width |>franka_poses_relative_to_camera |>object_id |>score_analytical |>score_simulation |>score_wrench |>suctioncup |>contact_poses_relative_to_camera |>object_id |>score_analytical |>score_simulation |>suction_poses_relative_to_camera |>objects |>bbox_loose |>categories |>poses_relative_to_camera
I downloaded version 2 of the dataset (Real) cited in the README. I browsed all the files in the dataset, and I can't seem to find the bounding boxes. Could you please provide any indication on where I could find the objects' bounding boxes?