fix: 交叉熵损失计算

ginka/generator/loss.py

@@ -405,6 +405,10 @@ class WGANGinkaLoss:
 class RNNGinkaLoss:
     def __init__(self, num_classes):
         self.num_classes = num_classes
+        weight = torch.ones(self.num_classes)
+        weight[0] = 0.3
+        weight[1] = 0.5
+        self.weight = weight
         pass
     
     def rnn_loss(self, fake, target):
@@ -412,8 +416,5 @@ class RNNGinkaLoss:
         fake: [B, C, H, W]
         target: [B, H, W]
         """
-        weight = torch.ones(self.num_classes)
-        weight[0] = 0.3
-        weight[1] = 0.5
         target = F.one_hot(target, num_classes=self.num_classes).float()
-        return F.cross_entropy(fake, target, label_smoothing=0.1, weight=weight)
+        return F.cross_entropy(fake, target, label_smoothing=0.1, weight=self.weight)

ginka/generator/rnn.py

@@ -205,7 +205,7 @@ class GinkaRNNModel(nn.Module):
                 map[:, y, x] = tile_id[:]
                 now_tile = tile_id if use_self else target_map[:, y, x].detach()
                 
-        return output_logits, map
+        return output_logits.permute(0, 3, 1, 2), map
 
 def print_memory(device, tag=""):
     if torch.cuda.is_available():

ginka/train_rnn.py

@@ -86,7 +86,7 @@ def train():
     optimizer_ginka = optim.AdamW(ginka_rnn.parameters(), lr=1e-4, weight_decay=1e-4)
     scheduler_ginka = optim.lr_scheduler.CosineAnnealingLR(optimizer_ginka, T_max=800, eta_min=1e-6)
 
-    criterion = RNNGinkaLoss()
+    criterion = RNNGinkaLoss(32)
 
     # 用于生成图片
     tile_dict = dict()
@@ -158,7 +158,6 @@ def train():
                     val_cond = batch["val_cond"].to(device)
                     target_map = batch["target_map"].to(device)
                     
-                    B, T = val_cond.shape
                     fake_logits, fake_map = ginka_rnn(val_cond, target_map, False)
 
                     val_loss_total += criterion.rnn_loss(fake_logits, target_map).detach()