TigDog splits

[eliabntt]

hi there, can you share the splits used for the tigdog dataset (esp the validation set) and which 36 images per category you used during training?

[chaneyddtt]

Hi @eliabntt, we use the splits following the previous work , and here is the randomly selected label list for training.

[eliabntt]

Thanks

[eliabntt]

Hi @chaneyddtt. Can you confirm that the actually labelled images for the various sets are as follow?

Train has 6112 annotations and 7350 images
Valid has 1317 annotations and 1486 images
Train 36 has 31 annotations and 32 images

I'm having difficulties reproducing your splits. I'm running the python code to split the tigdog dataset. That's giving me 10152 images as expected. Then I'm using the three npy files to filter the filenames based on the original folder sorted list. So if a idx is on one npy file, I keep the corresponding filename.

[eliabntt]

Perhaps: do you have the file name lists for each of those splits or a COCO-style json?

[chaneyddtt]

Hi @eliabntt, please refer to the image for how I generate the index for labeled images. Basically I randomly select images with 9 keypoints appearing in the image.