dogyoonlee / dp-nerf Goto Github PK

Hi, I encounter a problem trying to deblur my own data.
the training data is like below:

the novel view is like this:

I wonder what is the problem, thank you. I only change the source directory, my data is the llff format.

Hi teams, thanks for sharing your awesome work!

I want to train your model on real scenes data, but unfortunately, I cannot because of my GPU memory limit.

Though I can reduce batch size but I want to see the result from the model that is used in your paper, so would you like to share the what you trained if it is fine?

Thanks!

Dear Dogyoon Lee1，

I hope this message finds you well.

I have a question regarding the command:

python run_dpnerf.py --config ./configs/blurpool/tx_blurpool_dpnerf.txt --expname $dir_to_log --ft_path ./<basedir>/<expname>/200000.tar --render_only --render_test

``
Could you please clarify whether the output of this command is related to the evaluation of novel view synthesis or the deblurring evaluation? Based on the flags --render_only and `--render_test`, I understand that the script is rendering the test set without further optimization, but I'm unsure if this corresponds to deblurring results or new view synthesis.

Thank you in advance for your help!

Best regards,

[TRAIN] Iter: 600 Loss: 0.023193560540676117 PSNR: 19.362762451171875 TIME: 0h:0m:40.74s
[TRAIN] Iter: 800 Loss: 0.01994657889008522 PSNR: 20.115781784057617 TIME: 0h:0m:53.53s
[TRAIN] Iter: 1000 Loss: 0.018907170742750168 PSNR: 20.629796981811523 TIME: 0h:1m:6.02s
Traceback (most recent call last):
File "/DP-NeRF-main/run_dpnerf.py", line 753, in
train()
File "/DP-NeRF-main/run_dpnerf.py", line 639, in train
loss.backward()
File "/lcy_DP/lib/python3.9/site-packages/torch/_tensor.py", line 396, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "/miniconda3/envs/lcy_DP/lib/python3.9/site-packages/torch/autograd/init.py", line 173, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [512]], which is output 0 of LinalgVectorNormBackward0, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

训练到1000iter时报错，这个怎么解决呀.是pytorch版本问题吗。我的版本为pyorch==1.12.1

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1024]], which is output 0 of LinalgVectorNormBackward0, is at version 1; expected version 0 instead.

I train dp-nerf on the blurred LLFF data ( generated by applying blur kernel), and resize the input to 252x188. The results are like this.


It seems like that the main obj in the image disappears, are there any parameters need to adjust with resolution?

Hi! Thanks to your great work!

I noticed that there exists some modifed EntropyLoss in your codebase which is not being used. Does it work in your experiments? It would be great if you could share your thoughts on it.