Chushu

[chushu1225]

The issue of unexpected time dimension and radial trajectories are solved with several modifications to the code:

• Modify the write_on_epoch_end in main.py and save_reconstructions fucntion to ensure the slices are stacked into volume correctly. Also modify the predict_step to return time_frame for the final volume stacking.
• Modify how kspace and mask dimensions are handled in load_volume of class InferenceSliceDataset. Duplicate both kspace_volume and mask by 2 in time dimension for T1w/T2w/Blackblood. Delete the mask[None, :,:, None] part.
• Modify get_volume_shape_info to duplicate at time dimension as well. The shape here is saved before load_volume, so it won’t automatically change with the duplication in kspace volume and mask.

To run inference: python /common/lidxxlab/chushu/PromptMR-plus/configs/inference/pmr-plus/cmr25-task2-val_chushu.yaml
To run validation: sbatch scripts/CMR2025_debug_T1w.sh
load results and save as png: python load_results_new.py