Hand Pose annotations V1 vs V2

Q&A
1

I am interested in using as many 3d hand-pose annotations as possible and I am unsure if I should use V1 or V2 of your dataset.

According to the ‘stats’ in the visualizer, V1 has 333 videos with handposes and V2 has 235. But you mention in your release notes that V2 has far more annotations.

Are these annotations from different videos? So should I combine the annotations of V1 and V2?
Or are the annotations in V2 more dense?

Thanks for your help

2

Once you download annotations, you should have collectively ~441 takes with manual annotations and ~966 takes with automatic annotations. Let me know if you are seeing something else.

3

Apologies, those numbers also include test set takes which we are not releasing. The metrics for those can be computed using our EvalAI leaderboard. When you download handpose, you should see the following stats:

Train set: #Manual Annotation Takes: 286, #Automatic Annotation Takes: 624
Val set: #Manual Annotation Takes: 65, #Automatic Annotation Takes: 179