Thanks Wilton, i don’t have the script downloaded.
Is this the entire set of files that i need to copy to /opt/Autodesk/shared/python/ ?
Hi Wilton,
i downloaded your file successfully.
Thanks so much
I downloaded and extracted the original file as well.
I replaced
flameSimpleML.py
train.py
in
/opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/
in your version there is …/packages/python3.11
in the original there is
…/packages/bin/f2py
Do i copy the folder python3.11
here?
…/packages/python3.11
or here?
…/packages/bin/python3.11
or do in need to create a folder .lib and copy it here?
…/packages/.lib/python3.11
Hi everyone,
I’ve deleted flamesimpleml from /opt/Autodesk/shared/python/
I’ve run rescan python hooks.
However flamesimpleml is still showing up.
How do i remove it?
Thanks
Deepak
Hi,
Its running now,
However im getting these errors
[flame@vxfhost ~]$ /opt/Autodesk/python/2025.2.2/bin/python /opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/train.py /home/flame/Downloads/train/bw_dataset_2025JAN30_1319_EHBI
numpy version: 2.2.2
torch version: 2.5.1+cu121
source channels: 3
target channels: 3
Traceback (most recent call last):
File “/opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/train.py”, line 895, in
main()
File “/opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/train.py”, line 804, in main
output = model(source * 2 - 1)
^^^^^^^^^^^^^^^^^^^^^
File “/opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/packages/.lib/python3.11/site-packages/torch/nn/modules/module.py”, line 1736, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “/opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/packages/.lib/python3.11/site-packages/torch/nn/modules/module.py”, line 1747, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “/opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/models/multires_v001.py”, line 810, in forward
up7 = torch.cat([self.upsample7(x_multires6),x_multires3],axis=1)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 44 but got size 45 for tensor number 1 in the list.
[flame@vxfhost ~]$
How could I fix this?
thanks
Deepak
Hi guys!
I must admit I’ve got this tool a bit abandoned at the moment but if there is an interest its not a big deal to revive it.
Do someone would like to join developing?
I’m happy to add contributors on github so it is easier to move on
1 Like
I sent you a DM to explain better and try to resolve this issue.
Thanks Wilton
Have a nice day
1 Like
Hi everyone,
This is basically working now. There are no errors.
I need to train the data much longer.
For it to work on linux i installed the original file from Talosh and then copied Wiltons archive, path below
Following has to be done with Wilton’s file
replace the .py files and add the lib to this folder: /opt/Autodesk/shared/python/flameSimpleML.v0.0.3.linux/flameSimpleML/packages/.lib
Also all training images have to be of the same resolution.
Hope this helps.
Have a great weekend everyone.
Cheers
Deepak
1 Like
Hi everyone,
I’d just like to add.
I’ve been doing something fundamentally wrong. I was just randomly selecting the clips to export in any order. This was putting the wrong clips into the folders.
The order in which you select the clips in flame decides which clip goes into source and target folders. I assume if the wrong clip goes into each folder the training data and model created would be wrong.
First click the clip to treat ( source clip )
Second click the clip to refer to ( target clip )
@talosh @wiltonmts am i correct about this?
As I was getting the wrong results in my tests so far.
Cheers
Deepak
Hi one last thing.
How do i change the location of this folder /home/flame/flameSimpleML_models
to be in another location?
I would like to move this folder to my centralized location and would like all the models to be created there.
Cheers
deepak
A simpler way, without modifying the code, is to start the training and then stop it, copy the .pth file to a centralized folder of your choice, and restart the training using the --model path argument at the end, specifying the path to your moved file. This will allow you to continue training from that checkpoint.
Regarding the training order, it is important to select your source sequence and target sequence in the correct order. I usually add _source and _target labels to the timeline to easily identify the correct files
Thanks Wilton,
I finaly figured out that the selection order is important.
I’ll try this out, not so sure how to do it
restart the training using the --model path argument at the end, specifying the path to your moved file
Hi @talosh @wiltonmts
all my clips are 25fps.
After rendering the model, in the export folder the src is still 25fps but the render is 59.94fps NDF.
Can this be controlled in some way?
Thanks
Deepak
Media hub settings?