problems about data preprocess in CVC14_txt_to_numpy.txt
SunnyWuYang opened this issue · comments
SunnyWuYang commented
Thanks for your team providing such wonderful work!
I have some questions about the code in CVC14_txt_to_numpy.txt.
I noticed that in read_x1_x2_y1_y2_from_txt_annotations() function, the x2, y2 are resized to 470(original value is 512), however, in bbox_info_combine_aligned() function, they are resized again( multiple 1.087).
- What's the purpose of doing this?
- Do both the Kaist dataset and the CVC-14 dataset use this operation?
CalayZhou commented
Thanks for your attention! This is because the original size of images in CVC-14 is 640*471, the original size of images in Kaist is 640*512. To apply the same implementation, we resize 640*471 (CVC-14) to 640*512 (KAIST) .
SunnyWuYang commented
Thanks for your reply!