dvlab-research / PanopticFCN

Fully Convolutional Networks for Panoptic Segmentation (CVPR2021 Oral)
Apache License 2.0
391 stars 53 forks source link

difference between "FPN-Summed" and "Semantic FPN" #5

Closed valencebond closed 3 years ago

valencebond commented 3 years ago

What is the difference between "FPN-Summed" and "Semantic FPN" in table5, described in Section3.3? As far as I know, the high-resolution feature (1/4 scale of the input image) in Semantic FPN is also obtained by upsampling and element-wise summation

yanwei-li commented 3 years ago

Hi, the FPN-Summed represents that we upsample and sum the feature from FPN directly (without convs). And Semantic FPN includes convs in each upsample stage.

valencebond commented 3 years ago

thanks for your response