Question about predictions['images'] in vggt_to_colmap.py

[wangxl0385]

Thanks for this excellent repo!

I'm trying to generate a COLMAP format reconstruction result using vggt_to_colmap.py. Here's my question:

The size of my input images is: height=3845, width=2160. In the load_and_preprocess_images function, this size is changed to height=924, width=518, and then cropped to height=518, width=518. The cropped images are then used to reconstruct the 3D point cloud and other data. However, in the process_images function, the original image is directly resized to the height=518, width=518 dimensions, normalized, and assigned to predictions["images"]. I'm unsure if this is problematic.

Looking forward to clarification.

[hjh530]

Same issue.

[Winner-Nick]

Same issue.

[GopiRajuMatta]

Same issue.

[jytime]

Closing this issue as we have provided a new script demo_colmap.py