Glomap and GSplat Integration Issue - No Splats Created

[DuVogel87]

Description:

I am encountering an issue while trying to use Glomap with GSplat. The alignment in Glomap appears to be quite good, but when loading the results into GSplat via the simple_trainer (I also tried postshot, but it didn’t work either), no splats are created. Instead, the entire scene becomes a big mess with no visible result.

Here is a breakdown of what I have done so far:

• Glomap alignment seems to work fine as I can visually inspect the results.
• The simple_trainer and postshot are both not generating any splats, and the scene is messy with no usable results.
• I have checked the required directories, and everything seems to be there:
  • The sparse folder
  • The images
• I’ve also made a screenshot at an earlier phase of training to show you the issue. However, the issue did not improve during later stages of training.

Logs & Warnings:

/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/PIL/TiffImagePlugin.py:935: UserWarning: Corrupt EXIF data. Expecting to read 2 bytes but only got 0. 
  warnings.warn(str(msg))
/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/PIL/TiffImagePlugin.py:935: UserWarning: Corrupt EXIF data. Expecting to read 2 bytes but only got 0.
warnings.warn(str(msg))
/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/PIL/TiffImagePlugin.py:935: UserWarning: Corrupt EXIF data. Expecting to read 2 bytes but only got 0.
warnings.warn(str(msg))
0%|                                                 | 0/30000 [00:00<?, ?it/s]
/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/PIL/TiffImagePlugin.py:935: UserWarning: Corrupt EXIF data. Expecting to read 2 bytes but only got 0.
warnings.warn(str(msg))
loss=0.257| sh degree=0| :   0%|           | 91/30000 [00:22<2:24:05,  3.46it/s]
connection handler failed
Traceback (most recent call last):
File "/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/websockets/legacy/server.py", line 236, in handler
await self.ws_handler(self)
File "/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/viser/infra/_infra.py", line 327, in serve
await asyncio.gather(
File "/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/viser/infra/_infra.py", line 447, in _message_producer
outgoing = await window_generator.anext()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ubu/splatting/gsplat/venv/lib/python3.12/site-packages/viser/infra/_async_message_buffer.py", line 89, in window_generator
self.id_from_redundancy_key.pop(message.redundancy_key())
KeyError: 'BackgroundImageMessage'

Question:

• Why does the Glomap result not work with GSplat? I have followed all the necessary steps, and everything seems to be in place, yet I’m not getting any valid output.
• Is there anything missing or incorrect in my setup that could be causing this?

I’d appreciate any insights or guidance on how to resolve this issue. I'll also make sure to attach this issue in the repository.

Thanks in advance!

Image:

[DuVogel87]

[DuVogel87]

[DuVogel87]

Okay, when thing i figured it, that when i use a very small dataset with around 200 images and run glomap, i then can train a splat from it in gsplat (and even in postshot).

so one issue might be the datasize?

[poly2it]

There seem to be some incompatabilities between GLOMAP and COLMAP datasets. For another example, see: https://github.com/ArthurBrussee/brush/issues/119.