调整优化器学习率，增加训练轮数，修改类别权重，更新学习率调度器参数，启用调度器详细输出

config/lr_scheduler.yaml

@@ -3,9 +3,9 @@ scheduler_type: LINEAR_WARMUP_THEN_POLY_SCHEDULER
 # total_iters=epochs * 训练的图像数量
 kwargs: |
   {
-    "warmup_iters": 2904, 
-    "total_iters": 290400, 
+    "warmup_iters": 5808, 
+    "total_iters": 580800, 
     "warmup_ratio": 0.000001, 
-    "min_lr": 0., 
+    "min_lr": 0.00000, 
     "power": 1.
   }

config/optimizer.yaml

@@ -3,7 +3,7 @@ base_config:
   optim_type: AdamW
   kwargs: |
     {
-      "lr": 0.00006,
+      "lr": 0.006,
       "weight_decay": 0.01,
       "betas": (0.9, 0.999)
     }

config/train.yaml

@@ -3,14 +3,14 @@ batch_size: 16
 image_height: 200
 image_width: 200
 workers: 0
-epochs: 100
+epochs: 200
 
 # 每一类的占比权重，如果要让每一类的占比权重相同，为1.0即可
 weight:
   - 1.0
-  - 4.0
+  - 3.0
   - 1.0
-  - 2.2
+  - 2.0
 
 # 数据集存放位置
 root:

learning_rate_scheduler.py

@@ -257,27 +257,27 @@ def get_lr_scheduler(optimizer: optim, scheduler_type: SchedulerType, kwargs=Non
                 optimizer=optimizer,
                 step_size=30,
                 gamma=0.1,
-                verbose=False
+                verbose=True
             )
         elif scheduler_type == SchedulerType.MULTI_STEP_SCHEDULER:
             return MultiStepScheduler(
                 optimizer=optimizer,
                 milestones=[30, 60, 90],
                 gamma=0.1,
-                verbose=False
+                verbose=True
             )
         elif scheduler_type == SchedulerType.EXPONENTIAL_SCHEDULER:
             return ExponentialScheduler(
                 optimizer=optimizer,
                 gamma=0.95,
-                verbose=False
+                verbose=True
             )
         elif scheduler_type == SchedulerType.COSINE_ANNEALING_SCHEDULER:
             return CosineAnnealingScheduler(
                 optimizer=optimizer,
                 t_max=5,
                 min_lr=0,
-                verbose=False
+                verbose=True
             )
         elif scheduler_type == SchedulerType.LINEAR_WARMUP_THEN_POLY_SCHEDULER:
             return LinearWarmupThenPolyScheduler(