nerfstudio icon indicating copy to clipboard operation
nerfstudio copied to clipboard

"ERROR: `input_path` is not a directory"

Open rockywind opened this issue 2 years ago • 3 comments

Describe the bug When I run the script in my own dataset, I met the error below! The script is that below. ns-process-data images --data /DATA2/xiaoquan.wang/nerfstudio/data/own_data/pic_tmp/img/ --output_dir /DATA2/xiaoquan.wang/nerfstudio/data/own_data/pic_tmp/ Screenshots image

rockywind avatar Jan 20 '23 15:01 rockywind

Does /DATA2/xiaoquan.wang/nerfstudio/data/own_data/pic_tmp/ exist and what files are in there?

tancik avatar Jan 20 '23 23:01 tancik

Hi, This is the data folder. image

rockywind avatar Jan 21 '23 12:01 rockywind

Hi, I modified these two codes so that they can run on the server. image

rockywind avatar Jan 21 '23 12:01 rockywind

new_data_0.zip This is my image file. I got same error.

[15:22:46] 🎉 Done downscaling images.                                                         process_data_utils.py:337
[15:22:48] 🎉 Done extracting COLMAP features.                                                       colmap_utils.py:137
[15:22:50] 🎉 Done matching COLMAP features.                                                         colmap_utils.py:151
[15:30:57] 🎉 Done COLMAP bundle adjustment.                                                         colmap_utils.py:173
──────────────────────────────────────────────  💀 💀 💀 ERROR 💀 💀 💀  ───────────────────────────────────────────────
Error running command: colmap bundle_adjuster --input_path 
/home/piraveen/Desktop/ns/data/small_outputs/colmap/sparse/0 --output_path 
/home/piraveen/Desktop/ns/data/small_outputs/colmap/sparse/0 --BundleAdjustment.refine_principal_point 1
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
ERROR: `input_path` is not a directory

Versions COLMAP 3.6 Ubuntu 20.04

pira998 avatar Mar 13 '23 11:03 pira998

I have also encountered this problem, this is because the number of images you provided is not enough to predict the camera poses. Take more photos or record a video instead

cplusx avatar Apr 15 '23 04:04 cplusx

It may also be because your png/jpg are not the correct pixel format. For instance, I used ffmpeg to subsample a video and split it into 5 separate groups (couldn't use ns-process-data video for this), but the default ffmpeg datatype is yuv420p I think. COLMAP / OPEN CV cannot read this pixel data type.

Anyway, when I changed this command:

ffmpeg -i ${1} "${outdir}/frames/%03d.png"

to this one:

ffmpeg -i ${1} -pix_fmt bgr8 "${outdir}/frames/%03d.png"

Then COLMAP worked.

Hope this can help other ppl as well!

mkim-MASI avatar Nov 15 '23 23:11 mkim-MASI