Hi authors, thanks for your great work! I would like to ask about the stmt training on mip360 datasets. Your paper mentioned: "we train models on data downsampled by a factor of 8 and rendered at successively higher resolutions (1×, 2×, 4×, and 8×)".
However, I found that this training file still uses {scene}/images for training (when using -r 4) since I printed the image_path in the readColmapCameras functions. In my understanding, we should use {scene}/images4 in this case. Is it correct?
Hi, it will resize the image to the target resolution for training. You can try to print the size of images during training. But yes, directly loading from imags_4 is more efficient.
Hi authors, thanks for your great work! I would like to ask about the stmt training on mip360 datasets. Your paper mentioned: "we train models on data downsampled by a factor of 8 and rendered at successively higher resolutions (1×, 2×, 4×, and 8×)".
However, I found that this training file still uses {scene}/images for training (when using -r 4) since I printed the image_path in the readColmapCameras functions. In my understanding, we should use {scene}/images4 in this case. Is it correct?