search

zyc00 / PartSLIP2

MIT License
29 stars 3 forks source link

hi, about the evaluation criteria! #5

Closed ljwwwiop closed 5 months ago

ljwwwiop commented 5 months ago

Hello! PartSLip and PartSLip++ are indeed excellent pieces of research. I would like to know if the paper is currently under review? Additionally, I am interested in whether the evaluation code provided is consistent with the metrics mentioned in the paper? I am also conducting research on zero-shot part segmentation and I am looking to cite PartSLip and PartSLip++ for performance comparison and analysis. Ensuring the fairness of the comparison is very important to me. I look forward to your reply.

ljwwwiop commented 5 months ago

image l see that, maintaining the same setting as in PartSLip, it means to randomly select 8 shapes from each of the 45 categories, is that correct? Does this imply that there might be variability in the results each time due to the randomness? Moreover, it seems that for zero-shot, only 45*8 level test samples are needed for comparison, right?

zyc00 commented 5 months ago

45 * 8 is used for few-shot fine-tuning, few-shot and zero-shot evaluation are both on the whole PartNet-Mobility, which is the test set of PartNet-E.

ljwwwiop commented 5 months ago

45 * 8 is used for few-shot fine-tuning, few-shot and zero-shot evaluation are both on the whole PartNet-Mobility, which is the test set of PartNet-E.

Thanks, I understand!