Some questions about the form of the dataset

[mmsx]

Hello author, this is a very meaningful work. I now have some of my own image datasets. I converted the image datasets into event datasets through the event camera simulator, but I don't know what the event datasets are in the code for training. What does a structure datasets look like?

[XiangZ-0]

Hello mmsx,
Sorry for the late reply. We have added the data preparation code Prepare_data.py, please refer to the README.md for detailed usage. We also provide an example data directory in Database/Raw. Hope this helps.
Thank you.

[mmsx]

Hello author, thank you so much for being able to provide the code for data preparation, which will be of great help to my experiment!

[mmsx]

Hello author, I use my own dataset to train the model, but because my GPU only has 24g of memory, resulting in me having to reduce my batch_size and num_frames value, and set the num_workers to 0, which causes my model to test the picture, the result is pure black image, when I use the same parameters to train the example image you gave, the effect is also very bad, do you know how to solve this problem？

[XiangZ-0]

Hello mmsx,
Have you tried randomly cropping the blurry images and event frames for training? In our experiments, training EVDI with randomly cropped 64x64 image patches, batch_size=4, and num_frames=49 will occupy around 16 G of GPU memory, and the performance is ok.