Questions about data organization and a reshape error when training the Denoising Module of rDL-SIM

[werringwu]

Thank you for releasing rDL-SIM. I’m trying to reproduce the two-stage training pipeline.

After finishing train_SR_Inference_Module.py, I moved on to train_rDL_Denoising_Module.py. For the denoising stage, I organized the data as follows:

data_train/rDL-SIM/DN/Microtubules/
training/000000xx/1.tif ... 9.tif # 3 directions × 3 phases (size 128×128)
training_gt/000000xx.tif # single GT image (size 256×256)
validate/00000xxx/1.tif ... 9.tif
validate_gt/00000xxx.tif

However, I hit a reshape error in cal_modamp:
ValueError: cannot reshape array of size 65536 into shape (128,128,3,3)

From the code, cal_modamp expects a 3×3 stack (Ny×Nx×9), but I mistakenly passed the GT image (1×256×256). Could you please confirm:

In the denoising stage, should cal_modamp always take the 9 raw (or SR-upsampled) frames rather than the single GT image?

Is the recommended setting to estimate k0/modulation amplitudes at the LR grid (128×128), i.e., pParam.Nx=Ny=128, while the GT remains 256×256 (scale=2)?

Are there any additional requirements on the file names of the 9 frames (e.g., strictly 1.tif…9.tif)?

Thanks a lot for your help and for the great work!

Best regards,