search

voldemortX / pytorch-auto-drive

PytorchAutoDrive: Segmentation models (ERFNet, ENet, DeepLab, FCN...) and Lane detection models (SCNN, RESA, LSTR, LaneATT, BézierLaneNet...) based on PyTorch with fast training, visualization, benchmarking & deployment help
BSD 3-Clause "New" or "Revised" License
851 stars 138 forks source link

The difference or significance between Feature Flip Fusion and the middle flip in data augmentation #179

Open zhuyongxiang1 opened 1 week ago

zhuyongxiang1 commented 1 week ago

The difference or significance between Feature Flip Fusion and the middle flip in data augmentation ？

voldemortX commented 1 week ago

The difference or significance between Feature Flip Fusion and the middle flip in data augmentation ？

Firstly, Feature flip fusion is not a very significant module. It seems to help the performance but is not very solid grounded.

Flip aug only modifies the data distribution, while the fusion module is a feature enhancement, they are two seperate things.