cherubicxn/neat

NEAT: Distilling 3D Wireframes from Neural Attraction Fields (CVPR 2024)

neat's Issues

Reconstruction from custom data

Hello to the author,

Thanks for the amazing work. I have been following your work on wireframe parsing. Is it possible to run the 3D line reconstruction on custom data? Suppose my input is wireframe parsing results for image in each view, and the camera parameters are known.

Best

Clean up the code

CUDA out of memory

Hello

Thanks for your amazing work. I'm currently trying to replicate the results on the abc dataset and while running the the second step of neat-final-parsing, I run into a CUDA out of Memory error as shown below:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 196.00 MiB (GPU 0; 15.77 GiB total capacity; 14.16 GiB already allocated; 150.12 MiB free; 14.57 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

I'm using p3.2xlarge EC2 instance which has a Tesla V100 GPU with 16 GB GPU memory. Could you mention the specifications of the compute used for your experiments?

Also, please let me know if the out of memory issue can be managed by changing any parameter in confs like num_pixels, split_n_pixels etc?

Segmentation fault (core dumped) : Visualization

Hello @cherubicXN
The visualization script in the repo is showing the following error:

[Open3D WARNING] GLFW Error: X11: The DISPLAY environment variable is missing [Open3D WARNING] Failed to initialize GLFW Segmentation fault (core dumped)

Can you help me with this? Is this an issue with open3D or something else?

How to evaluate the performance of NEAT?

Dear author,

Thanks for your great work.

I am a little bit confused with the evaluation metrics for ACC-J and ACC-L. Although I also noticed that you have kindly explained the details of evaluation metrics in the Miscellaneous part, I cannot understand how to evaluate junctions and lines using point clouds sampled from the predictions.

Thanks for your help.