search

zju3dv / NeuralRecon

Code for "NeuralRecon: Real-Time Coherent 3D Reconstruction from Monocular Video", CVPR 2021 oral
https://zju3dv.github.io/neuralrecon/
Apache License 2.0
2.03k stars 294 forks source link

target: no valid voxel when computing loss #116

Closed MarklearnML closed 1 year ago

MarklearnML commented 1 year ago

Hi,I got same troubles when training, just as the follow. Why I will get this? I guess it come from the full_tsdf_layer.npz. Because I find that my full_tsdf_layer.npz's values is all 1. But why?

请原谅我英语不太好，作者您好，我训练的时候出现了下面的问题 说是没有target的值 我猜是generate_gt 生成的tsdf有点问题 因为我发现all_full_9里的full_tsdf_layer里面的值全为1 我不太清楚这个正不正常 如果不正常 那么又是什么原因呢 我觉得应该是fusion.py里的出现了点问题 但我并没有修改里面的源码 作者看到了能帮忙解答一下吗 非常感激!!!

2022-10-14 19:52:10.801 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:11.161 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:11.555 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:12.057 | INFO | main:train:212 - Epoch 8/991, Iter 0/1632, train loss = 0.000, time = 19.652 2022-10-14 19:52:12.429 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:12.531 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:12.603 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:13.011 | INFO | main:train:212 - Epoch 8/991, Iter 1/1632, train loss = 0.000, time = 0.765 2022-10-14 19:52:13.202 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:13.355 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:13.500 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:13.888 | INFO | main:train:212 - Epoch 8/991, Iter 2/1632, train loss = 0.000, time = 0.867 2022-10-14 19:52:14.118 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:14.211 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:14.280 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:14.678 | INFO | main:train:212 - Epoch 8/991, Iter 3/1632, train loss = 0.000, time = 0.737 2022-10-14 19:52:14.880 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:15.005 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:15.073 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:15.453 | INFO | main:train:212 - Epoch 8/991, Iter 4/1632, train loss = 0.000, time = 0.759 2022-10-14 19:52:15.640 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:15.730 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:15.791 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:16.166 | INFO | main:train:212 - Epoch 8/991, Iter 5/1632, train loss = 0.000, time = 0.692 2022-10-14 19:52:16.338 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:16.432 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:16.514 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:16.885 | INFO | main:train:212 - Epoch 8/991, Iter 6/1632, train loss = 0.000, time = 0.712 2022-10-14 19:52:17.059 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:17.155 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:17.220 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:17.601 | INFO | main:train:212 - Epoch 8/991, Iter 7/1632, train loss = 0.000, time = 0.708 2022-10-14 19:52:17.782 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:17.892 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:18.038 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:18.467 | INFO | main:train:212 - Epoch 8/991, Iter 8/1632, train loss = 0.000, time = 0.856 2022-10-14 19:52:18.847 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:19.085 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss 2022-10-14 19:52:19.218 | WARNING | models.neucon_network:compute_loss:242 - target: no valid voxel when computing loss

MarklearnML commented 1 year ago

I solve the problem!

ktiays commented 1 year ago

I solve the problem!

I also met the same problem. How did you solve it?

everythoughthelps commented 1 year ago

I also met the same problem, how did you solve it?

everythoughthelps commented 1 year ago

I solve the problem!

I also met the same problem. How did you solve it?

did you solve the problem?

MarklearnML commented 1 year ago

I also met the same problem, how did you solve it?

Are you using the complete ScanNet dataset or the 5G mini dataset? It seems that the author's source code is based on the complete dataset. If you're using the 5G dataset, you may encounter errors and need to modify the file names.

everythoughthelps commented 1 year ago

谢了老哥，我用的那个袖珍数据集，要改好多bug

Sent from my iPhone

------------------ Original ------------------ From: GuoyuHuang @.> Date: Tue,Aug 29,2023 4:15 PM To: zju3dv/NeuralRecon @.> Cc: Pan Meng @.>, Comment @.> Subject: Re: [zju3dv/NeuralRecon] target: no valid voxel when computing loss (Issue #116)

I also met the same problem, how did you solve it? Are you using the complete ScanNet dataset or the 5G mini dataset? It seems that the author's source code is based on the complete dataset. If you're using the 5G dataset, you may encounter errors and need to modify the file names.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>