city-super / MatrixCity

Apache License 2.0
217 stars 9 forks source link

关于big city的street部分的pose文件生成问题 #38

Closed HarshWinterBytes closed 1 month ago

HarshWinterBytes commented 2 months ago

您好! 我尝试用generate_split.py生成big city的street部分的transform_pose.json文件。我注意到你们用max_x,min_x,max_y,min_y来过滤,我在使用时使用的是默认值,并统计了一下过滤前后的数据量(bottom、left、right、top四个区域一起),如下: 过滤前:train: 258015 frames, test: 28430 frames 过滤后:train: 6745 frames, test: 740 frames

由于过滤前后的数据量差异太大了,我想问一下是否是正常的,或者说对于big city的street部分不能使用默认的max_x,min_x,max_y,min_y?

会产生这个疑问的另一个原因是,我统计了一下你们提供的small city中street部分的pose文件中的max_x,min_x,max_y,min_y,发现值分别为:

max_x = -1.2021224975585938
min_x = -8.515166015625
max_y = 2.1065020751953125
min_y = -5.469083251953125

generate_split.py中的默认值有较大差异

期待您的回复。 祝好!

yixuanli98 commented 2 months ago

The default value is for the block A of street in small city, which is much smaller than bit city. You can define your own range.

HarshWinterBytes commented 2 months ago

The default value is for the block A of street in small city, which is much smaller than bit city. You can define your own range. 感谢您的回复!

请问有可以参考的值吗?我看数据中有很多无效的图片(只有地面或者相机朝向海面的),我要如何设置一个合理的范围来获得尽可能多的合理有效的数据呢?