Zhengxinyang
Zhengxinyang
same issue
You can use our function `generate_image_for_fid` to generate rendered images for your model. For Shading-image-based FID computation, we use [clean-fid](https://github.com/GaParmar/clean-fid), which supports to compare FID between two image folders. You...
Hi @hummat, we actually compute FID scores for each view and then average across all views.
As described in the paper, we use a ResNet to map the input image into the hidden space of the generator on the airplane category. This part of the code...
Sorry, I may have forgotten to mention some details. You should set `no_fix_normal=False` in `render_mesh` at utils/render/render.py when rendering original meshes from ShapeNet
To calculate 1-NNA, you can refer to [PointFlow](https://github.com/stevenygd/PointFlow). To calculate ECD, you can refer to [Octree-Transformer](https://github.com/GregorKobsik/Octree-Transformer).
Thanks for creating such an impressive demo. Although, for optimal results, the level of the marching cubes should be set to 0. I apologize for my delayed response.
It looks like there is a version conflict between skimage and numpy. I'm sorry I haven't run the code on colab, maybe someone else has a solution?
For each stage, it takes around 3 days for training on 8 V100 Nvidia GPUs.
What do you mean by mismatch?