📂 What's the difference between the provided image and depth files?

[Victorlouisdg]

1. Input Files

Among the files provided in the dataset, could you please confirm if the following files will be given as input files to us during the competition?

• image_left.png, depth_image.jpg, point_cloud.ply

Answer: Yes, you will receive all files in the observation_start folder.

Additionally, I would like to inquire about the difference between image_left.png and image_right.png.

Answer: The camera used for the competition is a Zed2i Stereo RGBD camera. It has two RGB cameras (left and right) that are 12 cm apart. The ZED SDK uses the slight difference in perspective between the left and right view to estimate the depth of the viewed scene. Importantly, the depth map is aligned to the left view. So if you only use one RGB view, it is simplest to use the left. However, participants are free to use the right RGB image as well, for example to run your own depth estimation.

Will both depth_image.jpg and depth_map.tiff be provided as inputs?

Answer: Yes both will be available, but note that there is an important difference between the two. The depth_image.jpg is created mainly for visualization, it is a quantized (i.e. integer values from 0-255) version of the depth_map.tiff. For your calculations, use the depth_map.tiff, as it contains the depth of the scene as floating point values in meters (e.g. 0.94 m, 1.52 m, etc. ).

2. Output file

Is it correct to submit grasp_pose.json as the output?

Answer: Yes, that's correct! Your only required output is a grasp pose, submitted in the same format as the provided grasp_pose.json files in the dataset.