OTS数据集结果较差

[snowy-land]

您好，我按照论文中的步骤在OTS数据集上进行训练后得到的结果与论文中的偏差较大，请问您能解答下我的疑问吗？以下是我两次的参数设置及训练结果：
learning_rate: 0.001
crop_size: [240, 240]
train_batch_size: 18
val_batch_size: 1
network_height: 3
network_width: 6
num_dense_layer: 4
growth_rate: 16
lambda_loss: 0.04
category: outdoor

Date: 2024-03-03 15:36:53s, Time_Cost: 1117s, Epoch: [1/20], Train_PSNR: 26.60, Val_PSNR: 25.39, Val_SSIM: 0.9442
Date: 2024-03-03 15:55:27s, Time_Cost: 1114s, Epoch: [2/20], Train_PSNR: 27.94, Val_PSNR: 25.84, Val_SSIM: 0.9521
Date: 2024-03-03 16:13:53s, Time_Cost: 1106s, Epoch: [3/20], Train_PSNR: 28.63, Val_PSNR: 26.28, Val_SSIM: 0.9470
Date: 2024-03-03 16:32:27s, Time_Cost: 1113s, Epoch: [4/20], Train_PSNR: 28.81, Val_PSNR: 24.46, Val_SSIM: 0.9313
Date: 2024-03-03 16:50:59s, Time_Cost: 1112s, Epoch: [5/20], Train_PSNR: 29.18, Val_PSNR: 24.72, Val_SSIM: 0.9414
Date: 2024-03-03 17:09:37s, Time_Cost: 1119s, Epoch: [6/20], Train_PSNR: 29.26, Val_PSNR: 26.37, Val_SSIM: 0.9504
Date: 2024-03-03 17:28:18s, Time_Cost: 1120s, Epoch: [7/20], Train_PSNR: 29.48, Val_PSNR: 27.17, Val_SSIM: 0.9569
Date: 2024-03-03 17:47:10s, Time_Cost: 1132s, Epoch: [8/20], Train_PSNR: 29.53, Val_PSNR: 25.83, Val_SSIM: 0.9480
Date: 2024-03-03 18:06:02s, Time_Cost: 1132s, Epoch: [9/20], Train_PSNR: 29.64, Val_PSNR: 26.42, Val_SSIM: 0.9520
Date: 2024-03-03 18:24:54s, Time_Cost: 1131s, Epoch: [10/20], Train_PSNR: 29.68, Val_PSNR: 26.63, Val_SSIM: 0.9548
Date: 2024-03-03 18:43:46s, Time_Cost: 1133s, Epoch: [11/20], Train_PSNR: 29.75, Val_PSNR: 26.73, Val_SSIM: 0.9528
Date: 2024-03-03 19:02:38s, Time_Cost: 1132s, Epoch: [12/20], Train_PSNR: 29.78, Val_PSNR: 26.67, Val_SSIM: 0.9537
Date: 2024-03-03 19:21:29s, Time_Cost: 1131s, Epoch: [13/20], Train_PSNR: 29.82, Val_PSNR: 26.42, Val_SSIM: 0.9519
Date: 2024-03-03 19:40:19s, Time_Cost: 1130s, Epoch: [14/20], Train_PSNR: 29.82, Val_PSNR: 26.25, Val_SSIM: 0.9504
Date: 2024-03-03 19:59:10s, Time_Cost: 1131s, Epoch: [15/20], Train_PSNR: 29.84, Val_PSNR: 26.49, Val_SSIM: 0.9520
Date: 2024-03-03 20:18:01s, Time_Cost: 1131s, Epoch: [16/20], Train_PSNR: 29.85, Val_PSNR: 26.55, Val_SSIM: 0.9522
Date: 2024-03-03 20:36:51s, Time_Cost: 1130s, Epoch: [17/20], Train_PSNR: 29.87, Val_PSNR: 26.46, Val_SSIM: 0.9508
Date: 2024-03-03 20:55:40s, Time_Cost: 1129s, Epoch: [18/20], Train_PSNR: 29.86, Val_PSNR: 26.45, Val_SSIM: 0.9505
Date: 2024-03-03 21:14:31s, Time_Cost: 1130s, Epoch: [19/20], Train_PSNR: 29.87, Val_PSNR: 26.50, Val_SSIM: 0.9522
Date: 2024-03-03 21:33:14s, Time_Cost: 1123s, Epoch: [20/20], Train_PSNR: 29.87, Val_PSNR: 26.47, Val_SSIM: 0.9515
——————————————————————————————————————
learning_rate: 0.001
crop_size: [240, 240]
train_batch_size: 24
val_batch_size: 1
network_height: 3
network_width: 6
num_dense_layer: 4
growth_rate: 16
lambda_loss: 0.04
category: outdoor

Date: 2024-03-26 01:35:30s, Time_Cost: 1131s, Epoch: [1/30], Train_PSNR: 28.54, Val_PSNR: 26.42, Val_SSIM: 0.9527
Date: 2024-03-26 01:54:25s, Time_Cost: 1134s, Epoch: [2/30], Train_PSNR: 28.74, Val_PSNR: 25.35, Val_SSIM: 0.9459
Date: 2024-03-26 02:13:19s, Time_Cost: 1134s, Epoch: [3/30], Train_PSNR: 29.08, Val_PSNR: 25.98, Val_SSIM: 0.9465
Date: 2024-03-26 02:32:13s, Time_Cost: 1134s, Epoch: [4/30], Train_PSNR: 29.15, Val_PSNR: 26.25, Val_SSIM: 0.9530
Date: 2024-03-26 02:51:02s, Time_Cost: 1129s, Epoch: [5/30], Train_PSNR: 29.34, Val_PSNR: 26.18, Val_SSIM: 0.9505
Date: 2024-03-26 03:10:00s, Time_Cost: 1138s, Epoch: [6/30], Train_PSNR: 29.40, Val_PSNR: 25.71, Val_SSIM: 0.9478
Date: 2024-03-26 03:28:56s, Time_Cost: 1135s, Epoch: [7/30], Train_PSNR: 29.51, Val_PSNR: 25.80, Val_SSIM: 0.9452
Date: 2024-03-26 03:47:52s, Time_Cost: 1136s, Epoch: [8/30], Train_PSNR: 29.55, Val_PSNR: 25.72, Val_SSIM: 0.9495
Date: 2024-03-26 04:06:46s, Time_Cost: 1134s, Epoch: [9/30], Train_PSNR: 29.60, Val_PSNR: 26.04, Val_SSIM: 0.9497
Date: 2024-03-26 04:25:37s, Time_Cost: 1131s, Epoch: [10/30], Train_PSNR: 29.63, Val_PSNR: 25.97, Val_SSIM: 0.9502
Date: 2024-03-26 04:44:31s, Time_Cost: 1134s, Epoch: [11/30], Train_PSNR: 29.68, Val_PSNR: 25.94, Val_SSIM: 0.9474
Date: 2024-03-26 05:03:23s, Time_Cost: 1133s, Epoch: [12/30], Train_PSNR: 29.68, Val_PSNR: 25.63, Val_SSIM: 0.9461
Date: 2024-03-26 05:22:16s, Time_Cost: 1133s, Epoch: [13/30], Train_PSNR: 29.71, Val_PSNR: 26.05, Val_SSIM: 0.9473
Date: 2024-03-26 05:41:08s, Time_Cost: 1132s, Epoch: [14/30], Train_PSNR: 29.72, Val_PSNR: 26.05, Val_SSIM: 0.9490
Date: 2024-03-26 06:00:05s, Time_Cost: 1137s, Epoch: [15/30], Train_PSNR: 29.72, Val_PSNR: 25.90, Val_SSIM: 0.9481
Date: 2024-03-26 06:19:02s, Time_Cost: 1138s, Epoch: [16/30], Train_PSNR: 29.73, Val_PSNR: 26.08, Val_SSIM: 0.9492
Date: 2024-03-26 06:37:51s, Time_Cost: 1129s, Epoch: [17/30], Train_PSNR: 29.72, Val_PSNR: 25.88, Val_SSIM: 0.9470
Date: 2024-03-26 06:56:47s, Time_Cost: 1136s, Epoch: [18/30], Train_PSNR: 29.73, Val_PSNR: 26.03, Val_SSIM: 0.9488
Date: 2024-03-26 07:15:45s, Time_Cost: 1138s, Epoch: [19/30], Train_PSNR: 29.74, Val_PSNR: 25.96, Val_SSIM: 0.9482
Date: 2024-03-26 07:34:36s, Time_Cost: 1132s, Epoch: [20/30], Train_PSNR: 29.74, Val_PSNR: 25.93, Val_SSIM: 0.9482
Date: 2024-03-26 07:53:24s, Time_Cost: 1128s, Epoch: [21/30], Train_PSNR: 29.74, Val_PSNR: 26.02, Val_SSIM: 0.9486
Date: 2024-03-26 08:12:09s, Time_Cost: 1125s, Epoch: [22/30], Train_PSNR: 29.73, Val_PSNR: 25.96, Val_SSIM: 0.9481
Date: 2024-03-26 08:31:00s, Time_Cost: 1131s, Epoch: [23/30], Train_PSNR: 29.75, Val_PSNR: 26.04, Val_SSIM: 0.9487
Date: 2024-03-26 08:49:55s, Time_Cost: 1134s, Epoch: [24/30], Train_PSNR: 29.74, Val_PSNR: 26.02, Val_SSIM: 0.9487
Date: 2024-03-26 09:08:50s, Time_Cost: 1136s, Epoch: [25/30], Train_PSNR: 29.74, Val_PSNR: 26.02, Val_SSIM: 0.9486