AHupuJR / EFNet

Event-based Fusion for Motion Deblurring with Cross-modal Attention (ECCV'22 Oral) https://ahupujr.github.io/EFNet/
Other
145 stars 16 forks source link

Inferring from custom data #23

Closed huangfeng95 closed 1 month ago

huangfeng95 commented 1 month ago

hi, Thank you for your excellent work! What I am more concerned about is whether it is necessary to retrain the network weights if using my own RGB-event data (captured through ColorDAVIS346) for deblurring testing? In fact, I used the two pre trained weight files you provided, but the results were not very satisfactory, although they were not problematic on your dataset. And I need to confirm that there are no issues with my data preprocessing process: using events_to_accumulate_voxel_torch in base_datasets.py to process the raw events into voxels, and using the mask generation operation in make_voxels_real.py. Hope for your help.