boundingbox error

[Ricky-PLUS]

{
"index": 7504,
"category": "Depth and Distance",
"subcategory": "Relative",
"question_type": "multi-choice",
"input_modality": "single-image",
"question": "Estimate the real-world distances between objects in this image. Which object is closer to the table (highlighted by a red box), the lamp (highlighted by a blue box) or the television (highlighted by a green box)?\n(A) lamp\n(B) television\nThe coordinates of bounding boxes are: red: [805.4117431641, 512.9778442383, 903.9285888672, 687.9032592773] blue: [144.779876709, 542.463684082, 256.5966796875, 639.5326538086] green: [478.4644165039, 668.9904785156, 625.9038085938, 750.4893798828]",
"answer": "(A)",
"img_paths": [
"./dataset/CV-Bench/img/3D/distance/omni3d_hypersim_9.jpg"
],
"source": "cvbench",
"index_origin": 2047,
"category_origin": "Distance"
},
For the bounding boxes in the data annotation of the cvbench dataset, the object description, color, and numerical values of each bounding box cannot correspond.For example, the value of the red bounding box actually corresponds to the blue bounding box, [805.4117431641, 512.9778442383, 903.9285888672, 687.9032592773] is the lamp. ？

[haoningwu3639]

Thanks for your reminder. We have double-checked this issue and found that there was an oversight: in the original CV-Bench dataset, there are some differences in the order of the bounding boxes. In the 3D-Depth task, the boxes correspond to the colors red, blue, and green; while in the 3D-Distance task, the boxes correspond to blue, green, and red, respectively. We have now fixed this issue and updated our data JSON files accordingly.

Building such a large-scale and comprehensive benchmark is labor-intensive, and it is inevitable that some minor errors may be introduced. We kindly ask for your understanding and patience.
If anyone notices any other potential data issues, please feel free to let us know for further discussion!

Thanks again for your contribution!