MemoryError

[ideszscu]

This is what I get when I try to optimize 308 pictures. I don´t know if they are to much, or should I scale it before or I don't have enough space disk... any idea? Thanks

Optimizing
Output folder: ./output/cec3bb9c-6 [24/10 09:40:09]
Tensorboard not available: not logging progress [24/10 09:40:09]
Reading camera 308/308 [24/10 09:40:11]
Loading Training Cameras [24/10 09:40:11]
[ INFO ] Encountered quite large input images (>1.6K pixels width), rescaling to 1.6K.
 If this is not desired, please explicitly specify '--resolution/-r' as 1 [24/10 09:40:11]
Traceback (most recent call last):
  File "train.py", line 216, in <module>
    training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations, args.checkpoint_iterations, args.start_checkpoint, args.debug_from)
  File "train.py", line 35, in training
    scene = Scene(dataset, gaussians)
  File "C:\Users\Usuario\pinokio\api\gaussian-splatting-Windows.git\scene\__init__.py", line 73, in __init__
    self.train_cameras[resolution_scale] = cameraList_from_camInfos(scene_info.train_cameras, resolution_scale, args)
  File "C:\Users\Usuario\pinokio\api\gaussian-splatting-Windows.git\utils\camera_utils.py", line 58, in cameraList_from_camInfos
    camera_list.append(loadCam(args, id, c, resolution_scale))
  File "C:\Users\Usuario\pinokio\api\gaussian-splatting-Windows.git\utils\camera_utils.py", line 41, in loadCam
    resized_image_rgb = PILtoTorch(cam_info.image, resolution)
  File "C:\Users\Usuario\pinokio\api\gaussian-splatting-Windows.git\utils\general_utils.py", line 22, in PILtoTorch
    resized_image_PIL = pil_image.resize(resolution)
  File "C:\Users\Usuario\pinokio\bin\miniconda\envs\gaussian-splatting\lib\site-packages\PIL\Image.py", line 1943, in resize
    return self._new(self.im.resize(size, resample, box))
MemoryError

[SegaMega99]

same problem.

[ltm7725]

Same problem here. Anything?

[barrykeenan]

I am also getting the same error on a high-spec cloud computer, but I am able to generate a successful splat on my laptop that has lower specs. Any help appreciated.

File "C:\ProgramData\Anaconda3\envs\gaussian_splatting\lib\site-packages\PIL\ImageFile.py", line 283, in load_prepare
    self.im = Image.core.new(self.mode, self.size)
MemoryError

Error is happening on a vagon cloud computer with the following specs:

• Windows Server 2022
• 4 cores
• 16GB RAM
• 24GB GPU A10G Tensor Core GPUs

But, I can successfully run python train.py -s data\my_project -r 8 on my laptop with the following specs:

• Windows 11 Home
• Intel Core i7
• 16GB RAM
• 8GB GPU NVIDIA GeForce RTX 3070 Laptop

The vagon cloud computer should have plenty of resources. Any idea why I would be getting a MemoryError?

Thanks, Barry

Update: I suspect this is due to a lack of virtual memory on the cloud computer. When train.py resizes the source images, it causes a huge spike in system RAM. Could the script be modified to resize the images first in batches, and write them to disk, before starting training?

[barrykeenanresn]

Related question about quality - I have 483 input images of resolution 5332 x 3522, so when running train.py I get the warning:

[ INFO ] Encountered quite large input images (>1.6K pixels width), rescaling to 1.6K.

If I rescale the input images first to 1600 wide, rather than with the train.py -r parameter, will there be any loss in quality of the resulting splat? i.e. does train.py use any extra information from the original size images?

[WeihongPan]

same problem here.