Closed jatentaki closed 1 year ago
Thanks for bring this up. I upload the 1000 point cloud here (if you want to directly use this data, please verify the point cloud's axis/pose align with your training data)
for the pre-processing scheme, I normalized both the sampled data and the validation data into [-1,1] using this part of code, i.e. I call the function with compute_score(samples='samples.pt', ref_name='ref_ns_val_all.pt', norm_box=True)
Hello, I'm bringing up the questions I had after you closed #16, in case you missed that part. I would be interested to know as much as possible about the (sub)set of examples you used to evaluate on ShapeNet-vol, such that I can meanigfully compare to LION in absence of released weights/samples. The easiest would be if you can share that subset as a files, but in absence I would need
Thanks for your swift responses on the previous issues.