关于training_example.json 复现

[MickeyFei]

我找到了示例训练数据在mind2web中的原始内容，但是坐标计算对不上，请问可以分享下坐标计算方法嘛？ 我理解training_example.json 的第一轮对话的坐标（495,30）是相对坐标，该条数据对应的图片分辨率是1280*720，计算出的绝对坐标是（633.6,21.6） 该条原始数据对应的第一个框的坐标是
"bbox": { "x": 595.25, "y": 0.0, "width": 118.15625, "height": 60.0 }, 框的中心点的绝对坐标为（654,30） 请问这样的误差是否正确呢，训练数据对框的处理是怎样的呢？

[JjjFangg]

All input data coordinates are normalized within the range of 0-1000 as relative coordinates, without involving absolute coordinate inputs.

[MickeyFei]

All input data coordinates are normalized within the range of 0-1000 as relative coordinates, without involving absolute coordinate inputs.

抱歉，说错了，是相对坐标，请问是如何处理得到框对应的的相对坐标呢？

[JjjFangg]

Consider an image with a resolution of 1920 × 1080 pixels, and a bounding box with absolute coordinates:

Corner	Absolute Coordinates (X, Y)
Top-left	(640, 270)
Bottom-right	(1280, 810)

Formula for Relative Coordinates

x' = (x / W) * 1000
y' = (y / H) * 1000

Where:

• x, y are the absolute coordinates.
• W, H are the image width and height.
• The multiplication by 1000 ensures normalized coordinates fall within [0, 1000].

Converted Relative Coordinates

Corner	Calculation	Relative Coordinates (X', Y')
Top-left	(640 / 1920) * 1000, (270 / 1080) * 1000	(333.33, 250.00)
Bottom-right	(1280 / 1920) * 1000, (810 / 1080) * 1000	(666.67, 750.00)

[zackschen]

Hi, sorry to bother you, but I've been following the algorithm you provided and I still can't figure out how you got the coordinates (495, 30).

@MickeyFei I was just wondering if the calculation method is clear to you.