search

yrcong / RelTR

RelTR: Relation Transformer for Scene Graph Generation: https://arxiv.org/abs/2201.11460v2
229 stars 45 forks source link

Some details about Open Image v6. #21

Closed JacobYuan7 closed 1 year ago

JacobYuan7 commented 1 year ago

Many thanks for the great work. Since the code for Open Image v6 is not release, may I ask about the data split you use for training? Do you use the train+val splits or only train split for training?

yrcong commented 1 year ago

Hi! Sorry for the late reply. We only use training images for the training.

But I guess there is no big difference between train+val and only train since the validation set is quite small.

JacobYuan7 commented 1 year ago

Hi! Sorry for the late reply. We only use training images for the training.

But I guess there is no big difference between train+val and only train since the validation set is quite small.

Many thanks for your reply! May I ask a follow-up question: is wmAP usually higher than mAP or lower? Would you mind giving some reasoning for this?

yrcong commented 1 year ago

I think wmAP should be higher usually. The models have better performance on the inference of high frequency relationships while these relationships have higher weights.

JacobYuan7 commented 1 year ago

I think wmAP should be higher usually. The models have better performance on the inference of high frequency relationships while these relationships have higher weights.

Sorry for the late reply. I agree with you. Many thanks for the information! I closed this issue as completed.