search

facebookresearch / maskrcnn-benchmark

Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.
MIT License
9.29k stars 2.5k forks source link

equation used in LevelMapper #871

Open shanggdlk opened 5 years ago

shanggdlk commented 5 years ago

❓ Questions and Help

The code leverages the equation: target_lvls = torch.floor(self.lvl0 + torch.log2(s / self.s0 + self.eps)) to find the corresponding feature level for each RoI. Here self.s0 by default is 224. While in original RPN paper, the authors said that they use 224 because 224 is the canonical ImageNet pre-training size. However, if we use a different dataset with different image size (e.g., COCO), then do I need to change 224 to another value?

Thanks!

alexwq100 commented 5 years ago

hi,@shanggdlk ,have you found the answer?