Open helq2612 opened 5 years ago
@helq2612 Thanks for your feedback. We'll check the annotations.
Have you fixed the annotations? @huanglianghua @got-10k
We have identified a few set of inconsistent annotations. We'll find a way to release the updated annotations soon upon fixed. Thanks.
I find the similar problem in GOT-10k_Train_008623 - 008637 and 009058 - 009059. And I also find that the contents of the first few images in these folders are not changed, they look like the same image. Maybe this is the reason that the annotations cannot be corresponded to the target in the images.
Hi, Have you corrected the wrong annotations in the training set? and thanks for the dataset and toolkit.
@huanglianghua Could you provide a list of incorrect videos.
Hi, Even I noticed that the Bounding Box coordinates are inconsistent. Some of them have height = 0 while the resolution of largest Bounding Box is greater than the resolution of image itself. Please let us know in case you have solved this issue. Our lab is interested in using your work. mention: @tyler-hayes
Hi, we have fixed all annotations and are doing the final proof-checking. The fixed annotations and data will be released soon, where around 0.4% of the annotations will be affected.
Is there any news?
Any news? This incorrectness would raise errors when we use inner-frames relationship in our training pipeline. We solve this problem by remove these error sequences from training data by this script at present.
Any news on this? According to the google drive link to download the dataset, there hasn't been any updates to the dataset since 2018.
Also @superkevingit the link to the script gives a 404. Could you please let me know where I can find the script now?
Hi, I checked the bounding boxes on each frame, and find some of them are outside of the image region. For example, in the video GOT-10k_Train_000399, at frame 66, the image size is (w=1920, h=1080), but the bounding box is [570. 477. 831. 607.], so that ymax = 477+607 = 1084 > 1080.
Also for the case that object moves outside of image, the bounding boxes are mostly annotated as [-1, -1, -1, -1], but for same cases, the bounding boxes are annotated as 0 in width or height. It would be nice if the bounding boxes are annotated consistently.