Closed fcakyon closed 2 years ago
Hi @fcakyon
Thank you for pointing out this error. I think the codes for this and anti-uav got a bit mixed up. Could you pull and check if it's working now? Thanks
Thanks a lot for the fast response @KostadinovShalon! When we shuffle the images and then split for train and val, then frames of the same video will be present in both training and validation set and it not an expected behaviour. Am I correct?
I'm shuffling the annotation files, which in the case of drone-vs-bird there's one file per video. So no, frames of the same video won't appear at both sets. That's what I was trying to do in the anti-uav dataset, which it's not necessary here, and that's why I think it got mixed up.
@KostadinovShalon you are right, thanks a lot!
@KostadinovShalon one last question, what is the seed you used for generating dronevsbird coco annotations? Because at the moment every time we run the given script, train/val split changes.
Hello,
Thanks a lot for this comprehensive repo! @KostadinovShalon @MattPoyser.
When I want to create coco annotation files using the image video and annotation folder, at this line: https://github.com/KostadinovShalon/UAVDetectionTrackingBenchmark/blob/c00cdf12086692ec02e43e07a0750db49d8f7baf/detection/utils/convert_drone_vs_bird_to_coco.py#L114
the value of
train_last_idx
is equal to 67547 but there are only 77 elements in_files
so test_files list is always empty: https://github.com/KostadinovShalon/UAVDetectionTrackingBenchmark/blob/c00cdf12086692ec02e43e07a0750db49d8f7baf/detection/utils/convert_drone_vs_bird_to_coco.py#L118Would you mind fixing this error and uploading the fully functional conversion script?
Bests