Giters

CAMMA-public / rendezvous

A transformer-inspired neural network for surgical action triplet recognition from laparoscopic videos.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Question about the pretrained module

bot-white-g1ve opened this issue · comments

commented

Hi, thanks for your great job.

But I've found an issue with the pretrained module's weight. In the essay, you have mentioned that 35 videos are used for training, 5 videos for validation and 10 for testing. However, when I use your pretrained weight to test 5 val videos, the accuracy reaches 90% which may be weird if these 5 videos haven't been used for training.

Are these released pretrained weights also generated from training on these 5 val videos?

commented

Dear user,

Please acquaint yourself with the dataset splits in https://arxiv.org/pdf/2204.05235 for the correct use of the dataset split and model weights.

Thanks