Update README.md

README.md

@@ -6,19 +6,19 @@ tags:
 ---
 # HSIMAE: A Unified Masked Autoencoder with large-scale pretraining for Hyperspectral Image Classification
 
-![image/png](https://cdn-uploads.huggingface.co/production/uploads/65979dfeb4b5c254cb8ed20e/vyn81w5ZnXWifmbX8By6T.png)
+![image/png](https://cdn-uploads.huggingface.co/production/uploads/65979dfeb4b5c254cb8ed20e/YjbxlXg5el3nySkcQkmq_.png)
 
-![image/png](https://cdn-uploads.huggingface.co/production/uploads/65979dfeb4b5c254cb8ed20e/ItHlquQYMlzo8wVMt_aVh.png)
+![image/png](https://cdn-uploads.huggingface.co/production/uploads/65979dfeb4b5c254cb8ed20e/dxLbojSBr4Kdt-cgA3su7.png)
 
 ## ✨ Highlights
-### Masked HSI Modeling with Large-Scale Pretraining
-The HSIMAE was pretrained by a large-scale HSI dataset, named HyspecNet-11k, then directly finetuned on four target classification datasets.
+### Large-Scale and Diverse Dataset for HSI Pretraining
+A large and diverse HSI dataset named HSIHybrid was curated for large-scale HSI pre-training. It consisted of 15 HSI datasets from different hyperspectral sensors. After splitting into image patches, a total of **4 million** HSI patches with a spatial size of 9×9 were obtained.
 
-### Multi-Scale PCA for Features Extract
-To address these distributional shifts caused by the different spectral resolutions and spectral ranges between hyperspectral sensors, a MS-PCA was used to extract the multi-scale features of HSI spectra and transform the raw spectra into fixed-length features.
+### New MAE Architecture for HSI domain
+A modified MAE named HSIMAE that utilized separate spatial-spectral encoders followed by fusion blocks to learn spatial correlation and spectral correlation of HSI data was proposed.
 
 ### Dual-branch finetuning to leverage unlabeled data of target dataset
-Dual-branch finetuning framework was proposed by using an extra unlabeled branch to further adapted the model to the distributions of the target dataset and suppressed the overfitting issue.
+A dual-branch fine-tuning framework was introduced to leverage the unlabeled data of the downstream HSI dataset and suppressed overfitting on small training samples.
 
 ## 🧑‍💻 Contact