Please tell me. I see that your data preprocessing is processed as uint8. I want to process lung data. The uint8 precision lung details will be lost. Is it okay if I save it as float32? After trying it, it seems that the GPU will overrun.

[yuanpengpeng]

Error reported during testing

Namespace(name='dif-net', epoch=400, dst_list='knee_cbct', split='test', combine='mlp', num_views=3, view_offset=0, out_res=256, eval_npoint=100000, visualize=False)
mixed_dataset: ['knee_cbct']
输出dst_name: knee_cbct
CBCT_dataset, name: knee_cbct, split: test, len: 1.
load ckpt from /mnt/d/谷歌下载/DIF-Net-main/DIF-Net-main/scripts/logs/dif-net/ep_400.pth
DIF_Net, mid_ch: 128, combine: mlp
Traceback (most recent call last):
File "/mnt/d/谷歌下载/DIF-Net-main/DIF-Net-main/code/evaluate.py", line 110, in
model.load_state_dict(ckpt)
File "/home/yuanpeng/conda/envs/python310Tigre/lib/python3.10/site-packages/torch/nn/modules/module.py", line 2152, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for DIF_Net:
size mismatch for view_mixer.layer.0.weight: copying a param with shape torch.Size([5, 10, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 3, 1, 1]).
size mismatch for view_mixer.layer.0.bias: copying a param with shape torch.Size([5]) from checkpoint, the shape in current model is torch.Size([1]).
size mismatch for view_mixer.layer.2.weight: copying a param with shape torch.Size([1, 5, 1, 1]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).

    May I ask what is the reason for this? I followed the same data preprocessing steps and ran it on the 3090ti server and 24g video memory.

[lyqun]

To convert the data to uint8, you should

1. normalize the value scale of the data to [0, 1];
2. multiply the data by a factor of 255;
3. convert the data type to uint8.

Float32 is fine, but more memory is required and the data loading is much slower.