search

seoungwugoh / RGMP

Fast Video Object Segmentation by Reference-Guided Mask Propagation
192 stars 37 forks source link

Accuracy without pre-training #12

Open 0316038 opened 4 years ago

0316038 commented 4 years ago

Hi, I would like to ask about the ablation study without pre-training(-PT). How many epoch did you train and how to set the learning rate for the fine-tuning part to get the accuracy in Table 5? (J mean is 68.6 and F mean is 68.9) Thanks for your attention and hope to hear your reply soon.

seoungwugoh commented 4 years ago

Hi, For the purpose of ablation study, we apply the same training schedule with the counter part (with pre-training). Please refer to our paper for details.

huhu5036 commented 4 years ago

Hi, Did you ever try to set the prior frame mask to all black? I would like to know how much the reference frame and mask matter? Thus, the method can really handle it well when the target appears again from disappeared.
Thanks a lot, and hope for your reply.

seoungwugoh commented 4 years ago

@huhu5036 We have the exact experiment in our paper (Fig.5 and Table.5). You can see the effect of the reference frame and the previous mask.