search

rhett-chen / graspness_implementation

My implementation of Graspnet Graspness.
Other
139 stars 33 forks source link

The difference between 'In Paper' and 'In Repo' #4

Closed WWSBH closed 2 years ago

WWSBH commented 2 years ago

Dear Author, I am so much interested in this project and find that there is much difference between results of 'In Paper' and 'In Repo', especially for the dataset of realsense. So I am curious about what kind of reasons that cause the results distinction. Also, I have trained the whole network but can not achieve the performance you mentioned in the paper(only 60 AP for Seen-Realsense). What else operation I need to do to reproduce the results(Table-2) in your paper?

rhett-chen commented 2 years ago

Hi. The results of 'In paper' is the results reported by the authors of graspness, they haven't released the code, so I implemented the model by myself and reported the results of 'In Repo'. As for the performance of the results you mentioned for Seen-Realsense , it may be caused by different GPUs, torch versions, etc.

WWSBH commented 2 years ago

Fine. Thanks for your reply!