dm-de
commented
7 months ago @iFimo I tested COLMAP using this tutorial and was able to create dense cloud with GUI: https://www.youtube.com/watch?v=mUDzWCuopBo But for the moment, I did not train yet. See also inside my linked post below - here are some hints for train Can you try COLMAP and post some feedback?
Can you also give me some hints?
1) Can you confirm, that "tt_experiments.zip" works? Here are no checkpoint's in zip. See here my post: https://github.com/lfranke/TRIPS/issues/22
2) What is your GPU memory usage with viewer.exe & playground scene I have a issue with viewer and I'm not sure, if this happen, because I have only 8GB VRAM or because my exe ist compiled wrong. I think this has something to do with saiga & cuda compute capability. My special problem is, that I compile on one machine and copy release to another machine (where I have no admin rights). For this I need to set cuda version manually... It is hard to find out what is wrong.
iFimo
lh-dm
lfranke
jonstephens85
Crush1111
First of all, some feedback.
After days of trial and error, @dm-de provided the final step to compile it with a running viewer. That saved me. Originally posted by @dm-de in https://github.com/lfranke/TRIPS/issues/15#issuecomment-1936175493
And it looks really amazing. Great work, @lfranke + team!
With the “new” instructions in the readme, everything works fine.
Anyway, I have two comments:
1: For Windows conda prompt shell, I can’t use
./build/bin/RelWithDebInfo/viewer.exe --scene_dir scenes/tt_train. I have to use the full path, because the conda prompt shell, e.g.,C:\Users\USERNAME\TRIPS\build\bin\RelWithDebInfo/viewer.exe --scene_dir scenes/tt_train.The same goes for the training command.
This could be added to the readme file, as well.
2: It’s a bit confusing to have a scene named
“train”and the training command is also called“train”. Maybe you could rename the whole“train” sceneto something like“locomotive”(in the readme also). So that it is clear when it is the"train" commandand when it is the"train" scene="locomotive" scene.Now the training.
I've created really cool stuff with 3D Gaussian Splatting, but I'm still failing with TRIPS at the moment.
I've tried a lot and tried to write everything here, but I think reading all of this will be more confusing and time-consuming than helpful.
It starts with the "scenes" folder and the "experiments" folder. I thought the “scenes” folder was for raw training data and the “experiments” folder was for training results. However, this doesn't seem right as the viewer only needs the "scenes" folder to work. So how do these two folders relate to each other and what is which one for?
And much more importantly:
Can someone explain step by step with commands for windows conda prompt shell what i have to do, to train with my own training data, i.e. my photos? Please start with colmap, because I can't get a "dense_point_cloud.ply" file there, which I seem to need. I get several .ply files, but no
"dense_point_cloud.ply". I have COLMAP-3.9.1-windows-cuda.If you are still interested in how I tried to start the training but failed. Click here.
--- First of all, I tried to train the playground scene, and this seems works fine. But I can't manage to prepare my own images correctly. I have Colmap "COLMAP-3.9.1-windows-cuda", where I use the `“Automatic reconstruction”`. I tried a scene with 22 pictures just to understand the workflow. My scene called `“audih3”` in this case. After automatic reconstruction in Colmap, I save the project to my ColmapScenes folder. Here’s a screenshot of what I get. PS: I work on Windows 10 (screenshot just made with Mac to see the file structure). Screenshot_1: Colmap output.  --- I put the ColmapScenes folder in my TRIPS root (makes things easier for me). --- Then I used the following commands in my TRIPS root folder: ``` C:\Users\USERNAME\TRIPS\build\bin\RelWithDebInfo\colmap2adop.exe --sparse_dir ColmapScenes\cm_audih3\sparse\0\ --image_dir ColmapScenes\cm_audih3\images\ --point_cloud_file ColmapScenes\cm_audih3\dense_point_cloud.ply --output_path scenes\tt_audih3 --scale_intrinsics 1 --render_scale 1 ``` What I am missing is the `“dense_point_cloud.ply”` in my Colmap output, the command is looking for? Shouldn´t it be there? Is it saved with another filename and do I need to rename any of the existing files? --- When I move on and ignore that there is no `“dense_point_cloud.ply”`, the command will create a `tt_audih3` folder in my `\Trips\scenes` and creates some data there. Screenshot_2: Created data in `scenes` folder  Here’s a list, which files I would expect, and which were created. ``` dataset.ini [required] Check Camera files [required] Check images.txt [required] Check camera_indices.txt [required] Check point_cloud.ply [required] Fail poses.txt [required] Fail masks.txt [optional] Fail exposure.txt [optional] Fail ``` (`source: https://github.com/lfranke/TRIPS/blob/main/scenes/README.md`) Why is there no `"point_cloud.ply"` and no `"poses.txt"`? Will this be created, when I have the `"dense_point_cloud.ply"` in my ColmapScenes folder? --- When I again ignore the missing files, do I then have to use: ``` C:\Users\USERNAME\TRIPS\build\bin\RelWithDebInfo\preprocess_pointcloud.exe --scene_path scenes/tt_audih3 --point_factor 2 ``` Will this create a `"point_cloud.bin"` out of my `"point_cloud.ply"` like it is in the pre-trained models? --- If everything is correct, I should then be able to execute the final command: ``` C:\Users\USERNAME\TRIPS\build\bin\RelWithDebInfo\train.exe --config configs/train_normalnet.ini --TrainParams.scene_names tt_train --TrainParams.name audih3 ``` Am I correct? Where are my mistakes?
I would be very thankful if someone could explain the training in detail.