Closed chensh1127 closed 2 weeks ago
Hi!
To train in "own data" mode, you need a folder containing only jpg, jpeg, or png image files. For example, a folder containing assets/example/*.jpg would become the custom dataset structure.
In this situation, the train command would be:
python train.py -s {your path}/assets/example --exp_name just_example --white_bg --iterations 1000 --own_data
In your case, if only images exist in ./data/coser/images
, it seems that running the command line as follows should work:
python train.py -s ./data/coser/images --exp_name test --white_bg --iterations 1000 --own_data
If you've been proceeding normally in this format, consider using an absolute path instead of a relative path for -s
. Also, if there's no dust3r checkpoint in dust3r/checkpoints/, it may not work. If the checkpoint is in a different location, add the additional option --dust3r_path {your path}/*.pth
.
Finally, if the video is not being saved, install an additional library by running pip install av
.
Please let me know if any additional problems occur.
Thanks, Jisang.
hey, you did a nice job.
I wonder before the train. what is the structure of my custom datasets?
I run the code:
python train.py -s ./data/coser --exp_name ./data/coser/images --white_bg --iterations 1000 --own_data
the error is :assert imgs, 'no images foud at '+root AssertionError: no images foud at
and my datasets is:
looking forward to your reply.