chenyu-xjtu / dualfluidnet

Repository of DualFluidNet: an Attention-based Dual-pipeline Network for Fluid Simulation.

DualFluidNet: an Attention-based Dual-pipeline Network for Fluid Simulation.

🔗 Paper link: https://arxiv.org/abs/2312.16867

🔥 The repository will be completed soon!! Please wait patiently...

Video demo:

dualfluidnet's Issues

Pre-trained models

Hi, I've been interested in your project lately.
Would it be possible for you to provide any pre-trained models associated with this project?

Question about Training Efficiency

作者您好，看了论文，对您的工作很感兴趣。
请问您有尝试过在多卡上进行训练么，我目前尝试训练该模型，但训练速度在0.12 ips左右，不确定是否为一个正常训练速度？
因此考虑用多卡加快训练速度。利用 torch.nn.DataParallel ，但在训练中出现错误：

File "/root/DeepLagrangianFluids/scripts/../models/default_torch.py", line 141, in compute_correction ans_conv = conv(inp_feats, pos, pos, filter_extent)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl 
result = self.forward(*input, **kwargs)
File "/root/miniconda3/lib/python3.8/site-packages/open3d/ml/torch/python/layers/convolutions.py", line 352, in forward
    out_features += self.bias
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!

方便的话可以提供训练的配置么。期待您的回复，谢谢！

Question about evaluation

作者您好！
我目前在尝试评估模型的Wasserstein distance指标，但使用代码中wasserstein_distance方法计算后得到的数值很小，在1e-4的尺度，而一般论文中在0.2左右。