Open IandRover opened 2 months ago
Hello, I also reproduce the code in the dataset Sen2_MTC_New using 4 RTX3090, The training loss is about 0.06 (mae metric) I have some question, I found the pixels of out images is negative, how to convert the mae metric to psnr, ssim metric Or, you can chat me, email: zhtzou@gmail.com
Hello, I also reproduce the code in the dataset Sen2_MTC_New using 4 RTX3090, The training loss is about 0.06 (mae metric) I have some question, I found the pixels of out images is negative, how to convert the mae metric to psnr, ssim metric Or, you can chat me, email: zhtzou@gmail.com
Hello. I've also reproduced this code by modifying the code based on @landRover's techniques. Specifically, I used the weight diffcr_new.pth from the folder pretrained. Here are the results I obtained:
mae | psnr | ssim | rmse |
---|---|---|---|
0.04189379885792732 | 26.9799747467041 | 0.8941709995269775 | 0.05102594569325447 |
Note that the author didn't provide the implementation of PSNR, SSIM, and RMSE, so I implemented it on my own. The results do not match those reported in the paper, which is very weird. I'm unsure if the discrepancy is due to my error.
I have not trained this model yet. So, I'm curious, have you trained this model? Also, have you got the test results using your training weights?
I've also sent a letter to your email. You can also drop a letter to liuyi2052697@foxmail.com.
Dear Authors,
Thank you for sharing your excellent paper and the complete codebase. We have a question regarding the reproduction and would greatly appreciate your guidance.
After following the instructions by running
python run.py -p test -c config/ours_sigmoid.json
and fixing several bugs, we observed an MAE of between 0.8674 and 0.9696 on the testing set, which is significantly higher than expected and reported. Could you kindly confirm if the code used for evaluation is accurate and whether the checkpoints are up-to-date?Below are the changes we made to ensure execution.
configs/ours_sigmoid.json
from pmaa to PMAA (where our data is)/models/network_x0_dpm_solver
.alpha_t = alpha_t[:,None, None,None]
andsigma_t = sigma_t[:,None, None,None]
incore/dpm_solver_pytorch.py
Line 292.We look forward to your feedback and thank you in advance for your assistance.
Best regards,