search

CUHK-AIM-Group / EndoGaussian

EndoGaussian: Real-time Gaussian Splatting for Dynamic Endoscopic Scene Reconstruction
https://yifliu3.github.io/EndoGaussian/
MIT License
100 stars 5 forks source link

I got images without rendering #17

Closed MichaelWangGo closed 4 months ago

MichaelWangGo commented 4 months ago

Hi authors,

Thanks for your excellent work!

I followed the instructions in README, but either for pulling or cutting, when I finished training the rendering results are black.

And when I tried to open the point cloud, it says NAN coords and 0 degenerated faces. pointcloud

Can you give me some suggestions?

yifliu3 commented 4 months ago

Hi, thanks for your attention.

I think the possible reason is rendered images are not converted to 0-255 froim 0-1 before saving, thus all saved values are 0. You can check this by printing the values right after the rendering. I'm wondering if you have followed render.py to get rendered images.

Besides, can you observe normal psnr during the training? If so, the problem should not come from the rendering but from others.

MichaelWangGo commented 4 months ago

Hi, thanks for your attention.

I think the possible reason is rendered images are not converted to 0-255 froim 0-1 before saving, thus all saved values are 0. You can check this by printing the values right after the rendering. I'm wondering if you have followed render.py to get rendered images.

Besides, can you observe normal psnr during the training? If so, the problem should not come from the rendering but from others.

Hi, thank you very much.

I print the rendering images, all pixels are 0., and yes I cannot see normal psnr during the training.

yifliu3 commented 4 months ago

That's strange. I'm wondering if you have strictly followed the intruction in README.md. By the way, we tested our codes on Linux system, and I'm not sure if there is any problems with other systems like Windows.

MichaelWangGo commented 4 months ago

That's strange. I'm wondering if you have strictly followed the intruction in README.md. By the way, we tested our codes on Linux system, and I'm not sure if there is any problems with other systems like Windows.

I run the code on linux system... but ok, let me double check the instructions. Thank you so much!

MichaelWangGo commented 4 months ago

Hi, I think the problem should be in the environment creation, the directory 'submodule/depth-diff-gaussian-rasterization/third_party' doesn't contain glm, so I git clone from here https://github.com/g-truc/glm/tree/5c46b9c07008ae65cb81ab79cd677ecc1934b903.

Except for this, I followed the instruction strictly.

yifliu3 commented 4 months ago

Hi. The machines that we use may not have the glm originally. I have faced similar problems with glm on my machine and used sudo apt-get install libglm-dev instead of cloning and installing from source. Perhaps the problem comes from this.

MichaelWangGo commented 4 months ago

Hi. The machines that we use may not have the glm originally. I have faced similar problems with glm on my machine and used sudo apt-get install libglm-dev instead of cloning and installing from source. Perhaps the problem comes from this.

That 's it! Problem solved after installing libglm-dev!

Thank you so much!!!

fusheng-ji commented 3 months ago

For who has no root authority to install libglm-dev with Linux machine, here is a practical solution: You can clone the Gaussian-rasterizer submodule from 4dgaussian repo by command below. git clone --recursive https://github.com/ingra14m/depth-diff-gaussian-rasterization.git pip install -e submodules/depth-diff-gaussian-rasterization Then the issue about point cloud and rendered images should be solved.