Mismatch in #images vs. annotations for certain sequences

[alrikai]

Hello,

I noticed that sequences airplane-14, airplane-15, and bus-4 seem to have an off-by-one mismatch in the number of images compared to the number of annotations (in groundtruth.txt, full_occlusion.txt, and out_of_view.txt). More specifically,

• airplane-14 has 2822 images; groundtruth.txt has 2821 lines.
• airplane-15 has 4501 images; groundtruth.txt has 4500 lines.
• bus-4 has 3972 images; groundtruth.txt has 3971 lines.

and the corresponding full_occlusion.txt and out_of_view.txt files have the same number of comma separated elements as the groundtruth.txt file for the above sequences.

I downloaded the dataset a few weeks ago (from the google drive link) at https://cis.temple.edu/lasot/download.html.

Am I the only one with this issue? Or is this issue already fixed in a more recent version of the dataset and I need to re-download it? (I would verify this myself, but it's a rather large download).

Thanks!

[HengLan]

Hi,

Sorry for the late reply. I was checking the sequences you mentioned and find that airplane-14 had 2821 images, airplane-15 has 4500 images, and bus-4 has 3791 images. How did you get the number of images for these three videos?

[HengLan]

You can use the hashing code for verifying. I checked again and the data is ok in the sever.

[alrikai]

Thanks for the verification, I had initially loaded the data in python and noticed the mismatch, then checked via the command line in the unzipped data set, i.e.

$ ls airplane/airplane-15/img/ | wc -l
4501

$ cat airplane/airplane-15/groundtruth.txt | wc -l
4500

I'll check the md5 hash, thanks!

[alrikai]

The hashes did not match, and after finally managing to re-download the dataset, it all seems good. Thanks for the help