search

Sharath-girish / efficientgaussian

Official implementation of "EAGLES: Efficient Accelerated 3D Gaussians with Lightweight EncodingS"
https://efficientgaussian.github.io/
Other
131 stars 6 forks source link

Unable to unpickle .pkl file #5

Open lurock12 opened 8 months ago

lurock12 commented 8 months ago

Hey, I am relatively new to the field of computer vision, so this may be a beginner's question: After running your code it outputs a .pkl file. I was not able to unpickle it. I was assuming to get a .ply file to view the model in SIBR or similar. How can I access the .ply file?

Thanks in advance.

Pixelgamer72 commented 5 months ago

Did you ever find out how to do it?

Pixelgamer72 commented 5 months ago

If you look into the train_eval.py file, there are the lines that save the finished set. If you find the ones with .compressed on with ctrl f there are lines above or below with # . remove the # and from them then add the # to the lines that say compressed. when you next run it saves as a .ply file

instead of scene.save_best.compressed() it would be scene.save_best() -> i assume removing the # is what you are meant to do, i just removed any compressed arguments from them, there are some other commented lines that I am not sure of the purpose like:

Sharath-girish commented 2 months ago

The .pkl file encodes the latents in a compressed format which is decoded via the function call in line 98 of scene/init.py, defined in line 430 of scene/gaussian_model_sq.py. The ply file can additionally be saved via the --save_ply argument during training which can be used for visualizing the point cloud on SIBR Viewer. Note that the point cloud in ply format will remain uncompressed.