🚀 segformer-b0-finetuned-morphpadver1-hgo-coord-v2
該模型是基於NICOPOI-9/segformer-b0-finetuned-morphpadver1-hgo-coord-v1在NICOPOI-9/morphpad_coord_hgo_512_4class數據集上進行微調的版本。它在評估集上取得了以下結果:
- 損失值:0.0408
- 平均交併比(Mean Iou):0.9952
- 平均準確率:0.9976
- 整體準確率:0.9976
- 準確率 0 - 0:0.9993
- 準確率 0 - 90:0.9958
- 準確率 90 - 0:0.9969
- 準確率 90 - 90:0.9983
- 交併比 0 - 0:0.9975
- 交併比 0 - 90:0.9929
- 交併比 90 - 0:0.9949
- 交併比 90 - 90:0.9955
📚 詳細文檔
訓練超參數
訓練過程中使用了以下超參數:
- 學習率(learning_rate):6e - 05
- 訓練批次大小(train_batch_size):1
- 評估批次大小(eval_batch_size):1
- 隨機種子(seed):42
- 優化器(optimizer):使用 OptimizerNames.ADAMW_TORCH,β值為(0.9, 0.999),ε值為 1e - 08,無額外優化器參數
- 學習率調度器類型(lr_scheduler_type):線性
- 訓練輪數(num_epochs):60
訓練結果
訓練損失 |
輪數 |
步數 |
驗證損失 |
平均交併比 |
平均準確率 |
整體準確率 |
準確率 0 - 0 |
準確率 0 - 90 |
準確率 90 - 0 |
準確率 90 - 90 |
交併比 0 - 0 |
交併比 0 - 90 |
交併比 90 - 0 |
交併比 90 - 90 |
0.0654 |
2.5445 |
4000 |
0.1134 |
0.9236 |
0.9604 |
0.9602 |
0.9729 |
0.9373 |
0.9642 |
0.9674 |
0.9265 |
0.9144 |
0.9233 |
0.9301 |
0.0552 |
5.0891 |
8000 |
0.1426 |
0.9161 |
0.9562 |
0.9561 |
0.9607 |
0.9538 |
0.9547 |
0.9555 |
0.9166 |
0.9146 |
0.9112 |
0.9218 |
0.0469 |
7.6336 |
12000 |
0.0633 |
0.9556 |
0.9774 |
0.9773 |
0.9811 |
0.9714 |
0.9744 |
0.9826 |
0.9588 |
0.9516 |
0.9545 |
0.9576 |
0.0378 |
10.1781 |
16000 |
0.0506 |
0.9650 |
0.9822 |
0.9822 |
0.9826 |
0.9773 |
0.9844 |
0.9844 |
0.9661 |
0.9601 |
0.9643 |
0.9696 |
0.0582 |
12.7226 |
20000 |
0.0402 |
0.9737 |
0.9867 |
0.9866 |
0.9925 |
0.9891 |
0.9791 |
0.9860 |
0.9774 |
0.9700 |
0.9699 |
0.9774 |
0.0322 |
15.2672 |
24000 |
0.0453 |
0.9707 |
0.9850 |
0.9851 |
0.9809 |
0.9843 |
0.9909 |
0.9840 |
0.9746 |
0.9715 |
0.9637 |
0.9728 |
0.0254 |
17.8117 |
28000 |
0.1030 |
0.9652 |
0.9823 |
0.9822 |
0.9895 |
0.9808 |
0.9748 |
0.9841 |
0.9761 |
0.9599 |
0.9583 |
0.9666 |
2.3028 |
20.3562 |
32000 |
0.0572 |
0.9745 |
0.9871 |
0.9870 |
0.9861 |
0.9839 |
0.9885 |
0.9896 |
0.9789 |
0.9717 |
0.9700 |
0.9773 |
0.0769 |
22.9008 |
36000 |
0.0225 |
0.9866 |
0.9932 |
0.9932 |
0.9960 |
0.9899 |
0.9939 |
0.9932 |
0.9893 |
0.9837 |
0.9849 |
0.9884 |
0.0512 |
25.4453 |
40000 |
0.0329 |
0.9850 |
0.9924 |
0.9924 |
0.9959 |
0.9867 |
0.9954 |
0.9917 |
0.9857 |
0.9820 |
0.9843 |
0.9878 |
0.3281 |
27.9898 |
44000 |
0.0301 |
0.9866 |
0.9933 |
0.9932 |
0.9958 |
0.9913 |
0.9907 |
0.9952 |
0.9899 |
0.9858 |
0.9843 |
0.9863 |
0.1536 |
30.5344 |
48000 |
0.0355 |
0.9889 |
0.9944 |
0.9944 |
0.9981 |
0.9927 |
0.9920 |
0.9949 |
0.9941 |
0.9855 |
0.9880 |
0.9880 |
0.0079 |
33.0789 |
52000 |
0.0256 |
0.9933 |
0.9966 |
0.9966 |
0.9979 |
0.9951 |
0.9961 |
0.9974 |
0.9956 |
0.9917 |
0.9934 |
0.9924 |
0.0074 |
35.6234 |
56000 |
0.0205 |
0.9938 |
0.9969 |
0.9969 |
0.9983 |
0.9970 |
0.9966 |
0.9956 |
0.9963 |
0.9923 |
0.9928 |
0.9939 |
0.0077 |
38.1679 |
60000 |
0.0255 |
0.9933 |
0.9967 |
0.9966 |
0.9985 |
0.9946 |
0.9964 |
0.9971 |
0.9954 |
0.9925 |
0.9919 |
0.9934 |
0.0061 |
40.7125 |
64000 |
0.0282 |
0.9945 |
0.9972 |
0.9972 |
0.9987 |
0.9958 |
0.9974 |
0.9969 |
0.9967 |
0.9916 |
0.9950 |
0.9945 |
0.0051 |
43.2570 |
68000 |
0.0262 |
0.9937 |
0.9969 |
0.9968 |
0.9987 |
0.9949 |
0.9959 |
0.9979 |
0.9968 |
0.9916 |
0.9934 |
0.9930 |
0.0047 |
45.8015 |
72000 |
0.0564 |
0.9912 |
0.9956 |
0.9956 |
0.9991 |
0.9950 |
0.9940 |
0.9943 |
0.9958 |
0.9882 |
0.9897 |
0.9912 |
0.0046 |
48.3461 |
76000 |
0.0492 |
0.9939 |
0.9969 |
0.9969 |
0.9992 |
0.9941 |
0.9974 |
0.9970 |
0.9969 |
0.9903 |
0.9938 |
0.9945 |
0.0552 |
50.8906 |
80000 |
0.0438 |
0.9948 |
0.9974 |
0.9974 |
0.9992 |
0.9966 |
0.9967 |
0.9972 |
0.9980 |
0.9924 |
0.9948 |
0.9941 |
0.0039 |
53.4351 |
84000 |
0.0361 |
0.9953 |
0.9976 |
0.9976 |
0.9991 |
0.9961 |
0.9973 |
0.9981 |
0.9975 |
0.9928 |
0.9952 |
0.9956 |
0.0034 |
55.9796 |
88000 |
0.0317 |
0.9958 |
0.9979 |
0.9979 |
0.9993 |
0.9964 |
0.9974 |
0.9985 |
0.9979 |
0.9937 |
0.9955 |
0.9963 |
0.0149 |
58.5242 |
92000 |
0.0408 |
0.9952 |
0.9976 |
0.9976 |
0.9993 |
0.9958 |
0.9969 |
0.9983 |
0.9975 |
0.9929 |
0.9949 |
0.9955 |
框架版本
- Transformers 4.48.3
- Pytorch 2.1.0
- Datasets 3.2.0
- Tokenizers 0.21.0
📄 許可證
許可證類型:other