search

Coordi777 / Conditional-Diffusion-for-SAR-to-Optical-Image-Translation

The official implementation of Conditional Diffusion for SAR to Optical Image Translation
19 stars 3 forks source link

results #5

Open szh404 opened 7 months ago

szh404 commented 7 months ago

Hey, @Coordi777
Regarding the test results, I have some doubts and hope you can provide some valuable suggestions.

I am using the SAR2Opt dataset, which consists of 1450 pairs of training samples.

1_0_0
1_0_0

I trained using the default parameters in the code you provided(Thank you).

However, when I sample, I find that the quality of the generated images is not good. Additionally, the values of SSIM and FID metrics are also poor.

model050000.pt 0 1

model070000.pt test_0.png 0 0 0

test_7.png 7 7 7

Is it because my training data amount is too small, leading to this phenomenon, or are there other settings reasons? I hope you can provide some advice.

ZhangJinian commented 7 months ago

Hi! May I ask what system you are running the code on? I tried running it on Windows but ran into difficulties.

szh404 commented 7 months ago

Hi! May I ask what system you are running the code on? I tried running it on Windows but ran into difficulties.

Ubuntu20.04， Pytorch1.11， Cuda11.3

ZhangJinian commented 7 months ago

Thank you！

Hi! May I ask what system you are running the code on? I tried running it on Windows but ran into difficulties.

Ubuntu20.04， Pytorch1.11， Cuda11.3

Thank you！

ZhangJinian commented 7 months ago

My training dataset has 32520pairs, and I have trained the model for 32950(model032950.pt.) My results are very bad. I think it’s because we don’t train it enough.

1711702862860

szh404 commented 7 months ago

My training dataset has 32520pairs, and I have trained the model for 32950(model032950.pt.) My results are very bad. I think it’s because we don’t train it enough.

1711702862860

Sorry for my late reply. I was busy with other things a few days ago and didn't reply promptly. How is the reproduction effect now? Is it the same as before? Regarding my SAR2OPT dataset, I trained for 100,000 iterations, and the performance of the model saved at 070000.pt was the best. However, it still didn't achieve the expected performance.

Best wishes!

879087734 commented 6 months ago

Brother, I would like to ask if you have succeeded in running and how it has worked

szh404 commented 6 months ago

Brother, I would like to ask if you have succeeded in running and how it has worked

Indeed, I has succeeded running the program, but the results wasn't satisfied and I don't know how to train (set hyper parameters) to get a good result. I'll carry on doing this.

879087734 commented 6 months ago

My code doesn't work, can we add contact information, I want to ask you some questions, are you Chinese? My QQ is 3032904759

Anurag-Mishra22 commented 2 months ago

@szh404 can you suggest how to get best output. Now are you satisfied with results. Plz guide.