πŸš€ Discussion on Starter Kit

Hi!

Link to the Starter Kit :point_left:

If you have any questions or issues about the Starter Kit, please do post them in this thread. If you want to discuss new ideas and strategies on how to go about the challenge, head over to πŸ’‘ Ideas Discussion Thread

Hi!
I noticed in the dataset that birds are named both β€˜BIrd’ and β€˜Bird’. Could be of interest if one make classes from the names.

There might be an error in the Dataset class when combining parts.
len(dataset.get_flight_ids()) => 3438, but there are 4154 flight sequences. The same flightid exist in more than one part, but the combined Dataset is probably limited to one of the sequences.

Thanks for a fun competition!

2 Likes

Hi @aanndd,

Regarding β€œbird question”
Yes, both BIrd and Bird refer to the same class - thanks for letting us know and apologies for inconvenience. We will fix that. Also not that birds are not considered encounters that should be detected - they are don’t care (are not counted for either detection or not).

Regarding β€œcombining parts”
Please note that train3 folder will have flight ids identical to train1 and train2. However, the corresponding videos will be different. Thus if you combing everything in one folder we suggest to concatenate folder name (e.g. train1) with flight_id.

Please let us know if you have any other questions.

1 Like

Thanks for the answer. When combining I was thinking about the starter kit Dataset class:

dataset**.add**(local_path, s3_path)

1 Like

Hi @aanndd, the Dataset class is updated for above edge case and should provide you all the 4154 flight sequences.

1 Like

6 posts were split to a new topic: Discussion on SiamMOT baseline

3 posts were merged into an existing topic: Discussion on SiamMOT baseline

2 posts were split to a new topic: Confusion on airborne object names and range_distance_m