Avalibilty of Dataset.\

[yl3800]

Thanks for the interesting work. Could you also share the dataset described in Sec IV. B, which supports Grasp Affordance Prediction on 3D AffordanceNet.

[sjauhri]

Hi @yl3800,
The working dataset generation and training code will be released a bit later (in August and September).
The grasp affordance prediction uses a simple extra MLP decoder head that classifies each grasp into each type of affordance.
The data for that is generated by remapping the 3DAffordanceNet data. See: https://github.com/iROSA-lab/3DAffordanceNet/blob/main/remap_filter_data.ipynb. Each grasp is associated with an affordance based on the proximity to the affordance pointcloud from the remapped 3DAffordanceNet

[yl3800]

Thx for the reply, may I ask where I can find the grasp annotation that you used for the filtered 3DAffordanceNet? or it will be released later? Thx!

[sjauhri]

Hi @yl3800,
Sorry for the late reply.
You can now find the method for affordance annotation of the grasps here: https://github.com/pearl-robot-lab/neugraspnet/blob/main/neugraspnet/scripts/data_gen/assign_grasp_affordance.py
ach grasp is associated with an affordance based on the proximity to the affordance pointcloud from the remapped 3DAffordanceNet