NaN or Inf found in input tensor

[xiaoqidaov2]

train_mouth这里报错是什么原因
tb_writer.add_images(config['name'] + "view{}_mouth/depth".format(viewpoint.image_name), (render_pkg["depth"] / render_pkg["depth"].max())[None], global_step=iteration)

/root/anaconda3/envs/TalkingGaussian/lib/python3.8/site-packages/tensorboardX/summary.py:286: RuntimeWarning: invalid value encountered in cast
tensor = (tensor * 255.0).astype(np.uint8)
NaN or Inf found in input tensor

debug后发现

render_pkg["depth"]
tensor([[[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.],
...,
[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.]]], device='cuda:0')

[xiaoqidaov2]

找到问题了，lms 存在大量负数