abhishek1411 / event-rgb-fusion Goto Github PK

In paper, the split of dataset is 30 sequences for training, 3 sequences validation, and 8 sequences testing. But in DSEC_detection_labels folder, the split of dataset is 33 sequences for training, 3 sequences validation, and 5 sequences testing. I am confused.

Hello, it is a wonderful job! I have a question about the code. Do we need to implement the alignment between event data and RGB data by ourselves?

Hello, this is a great job. Can I have a look at the paper? I can't download it in https://hal.archives-ouvertes.fr/hal-03591717/. I hope you can help me, thank you!

Hello,
Wonderful job! There is one question about the code:

These .txt files are written in binary. What is the content here, is it a picture? And I wonder if the contents be visualized?

Any advice will help me. Thanks!

Thanks for your great job!! I have a question about the misalignment between transformed_image and boxes.

[Situation]
I tried visualizing the object detection boxes on the image and noticed that all the boxes seem to be shifted about 10 pixels to the left. The alignment with the event data appears to be correct, so the homographic transformation seems to be working fine.

[Question]
Given this situation, I'd like to ask whether the inaccurate labeling could be the reason behind the misalignment of the object detection boxes or if there could be an issue with the visualization method I used.

I am wondering if you could kindly provide checkpoint files used directly for testing.Thanks a lot.

Great work!

Can you please provide the trained weights of the FPN-Fusion model?