initiate mae

_posts/computer_vision/papers/2024-09-02-mae.md

@@ -0,0 +1,16 @@
+---
+layout: post
+title: Masked Autoencoders Are Scalable Vision Learners
+category: Computer Vision
+tag: [mae]
+---
+
+# [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377)
+
+
+> People can misunderstand that this masked strategy training is just reconstructing the missing pixels in the image. However, since the missing masks are predicted without any ground truth of those missing pixels and by inducing the semantic information of the image without , this is not just reconstruction. That's why it provides better performance for fine-tuning or downsteram tasks, when pre-training the model with mae method.
+
+
+### References
+- [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377)
+- [code - github](https://github.com/facebookresearch/mae)