Update README.md

README.md

@@ -55,12 +55,9 @@ FAMA models were rigorously validated across three distinct transfer learning ta
 
 ### 1. Galaxy Classification (Full Fine-tuning)
 
-| Method | backbone | Pre-train Data | Accuracy on **galaxy-desi** (16k samples) | Accuracy on **galaxy-sdss** (Cross-Domain) |
-| :----- | :------- | :------------ | :------------------------------------------------ | :------------------------------------------------ |
+| Method | backbone | Pre-train Data | Acc on **galaxy-desi** | Acc on **galaxy-sdss**|
+| :----- | :------- | :------------ | :---------------------------------- | :-------------------------------- |
 | **FAMA (ours)** | ViT-H | DESI-1M | **89.10** | **96.02** |
-| **FAMA (ours)** | ViT-B | DESI-1M | 87.23 | 95.25 |
-| LVM | Swin | DESI-3.5M | 84.63 | N/A |
-| Scratch | ViT-B | None | 67.27 | 87.38 |
 
 ### 2. Gravitational Lensing Detection
 
@@ -69,9 +66,7 @@ FAMA achieves the highest Average Precision (AP) scores for strong gravitational
 | Method | backbone | AP | AP<sup>75</sup> |
 | :----- | :------- | :-- | :------------ |
 | **FAMA (ours)** | ViT-H | **42.62** | **49.43** |
-| **FAMA (ours)** | ViT-L | 41.31 | 47.04 |
-| LVM | Swin | 31.38 | 27.90 |
-| Scratch | Res-50 | 21.29 | 16.87 |
+
 
 ### 3. Redshift Prediction (Cross-Domain)
 
@@ -80,8 +75,7 @@ The pre-trained model on DESI data is fine-tuned on the SDSS Redshift dataset.
 | backbone | Δz (Bias, Lower is Better) | σ<sub>MAD</sub> (Dispersion, Lower is Better) |
 | :------- | :---------------------------- | :-------------------------------------------------- |
 | **FAMA ViT-H** | **0.51 × 10⁻⁴** | **0.56 × 10⁻²** |
-| FAMA ViT-B | 0.97 × 10⁻⁴ | 2.31 × 10⁻² |
-| Photo | 1.70 × 10⁻⁴ | 1.43 × 10⁻² |
+
 
 ## 🛠️ How to Use for Transfer Learning
 
@@ -108,7 +102,7 @@ The following fine-tuning configurations were used for the `galaxy-desi` classif
 
 | Config | ViT-Base | ViT-Large | ViT-Huge |
 | :----- | :------- | :--------- | :------- |
-| **Optimizer** | AdamW (β₁=0.9, β₂=0.999) | AdamW | AdamW |
+| **Optimizer** | AdamW | AdamW | AdamW |
 | **Learning Rate** | 1.5 × 10⁻³ | 2 × 10⁻³ | 1 × 10⁻³ |
 | **Batch Size** | 64 | 64 | 32 |
 | **TrainingEpochs** | 50 | 50 | 50 |