Unable to get the dataset from the paper
pandeydeep9 opened this issue · 0 comments
Dear authors,
Thank you for uploading the code to generate the MARVEL dataset.
I tried to obtain the dataset for vessel classification. I tried both VesselClassification.dat and VesselVerificationUpdated.dat.
I ran into a few problems.
First, I think many of the images may have been removed from Shipspotting website. After the download, I found many of the images to be missing for all 26 superclasses (e.g. for superclass 1 [ container ship], 93 training images and 37 test images were missing) It would be great if you can upload the original dataset or the missing instances.
Next, I tried to reorganize the dataset to the superclasses. I found that there are many repeated instances in the FINAL.dat file. For e.g. image 403220.jpg from superclass 20 ( 'Icebreaker' ) was repeated 3 times in FINAL.dat.
I found a similar repetition for imdb.mat file. In imdb.mat file, imdb.images.name has total 239622 instances but only 140243 of them are unique. Many of the rows in the file are repeated. For e.g ['D:\ShipspottingDataset\F2\2256472.jpg'] appears at index 579 and 920.
This leads to some unexpected behavior. For eg. for superclass 19, I get a total of only 1993 images (counting both training and test images) but I think the total number of images should be 9216 for all superclasses.
I wonder if this is expected or have I made some mistake?
Thanks,
Deep S. Pandey