diff --git a/ginka/train_vae.py b/ginka/train_vae.py
index e440ac8..53bbf9b 100644
--- a/ginka/train_vae.py
+++ b/ginka/train_vae.py
@@ -56,6 +56,8 @@ from shared.image import matrix_to_image_cv
 BATCH_SIZE = 128
 LATENT_DIM = 48
 KL_BETA = 0.1
+SELF_GATE = 0.5
+GATE_EPOCH = 5
 
 device = torch.device(
     "cuda:1" if torch.cuda.is_available()
@@ -101,6 +103,7 @@ def train():
     criterion = VAELoss()
     
     self_prob = 0
+    prob_epochs = 0
 
     # 用于生成图片
     tile_dict = dict()
@@ -153,16 +156,21 @@ def train():
         #     val_loss_total = torch.Tensor([0]).to(device)
         #     for batch in tqdm(dataloader_val, desc="Validating generator.", leave=False, disable=disable_tqdm):
         #         target_map = batch["target_map"].to(device)
-                
+        
         #         fake_logits, mu, logvar = vae(target_map, 1 - gt_prob)
 
         #         loss, reco_loss, kl_loss = criterion.vae_loss(fake_logits, target_map, mu, logvar, KL_BETA)
         #         val_loss_total += loss.detach()
         
         # avg_loss_val = val_loss_total.item() / len(dataloader_val)
+
         # 先使用训练集的损失值，因为过拟合比较严重，后续再想办法
-        if avg_loss < 0.5 and self_prob < 1:
+        if avg_loss < SELF_GATE:
+            gate_epochs += 1
+
+        if gate_epochs >= GATE_EPOCH and self_prob < 1:
             self_prob += 0.01
+            gate_epochs = 0
         
         scheduler_ginka.step(avg_loss, self_prob)