🚀 segformer-b0-finetuned-morphpadver1-hgo-coord-v3
該項目是一個圖像分割模型,基於nvidia/mit-b3
模型在特定數據集上微調而來,在評估集上取得了優異的成績,能夠高效準確地完成圖像分割任務。
🚀 快速開始
此模型是 nvidia/mit-b3 在 NICOPOI-9/morphpad_coord_hgo_512_4class 數據集上的微調版本。它在評估集上取得了以下結果:
- 損失值:0.0298
- 平均交併比(Mean Iou):0.9952
- 平均準確率:0.9976
- 總體準確率:0.9976
- 0 - 0 準確率:0.9983
- 0 - 90 準確率:0.9969
- 90 - 0 準確率:0.9980
- 90 - 90 準確率:0.9971
- 0 - 0 交併比:0.9959
- 0 - 90 交併比:0.9950
- 90 - 0 交併比:0.9945
- 90 - 90 交併比:0.9954
🔧 技術細節
訓練超參數
訓練過程中使用了以下超參數:
- 學習率:6e - 05
- 訓練批次大小:1
- 評估批次大小:1
- 隨機種子:42
- 優化器: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:線性
- 訓練輪數: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 |
總體準確率 |
0 - 0 準確率 |
0 - 90 準確率 |
90 - 0 準確率 |
90 - 90 準確率 |
0 - 0 交併比 |
0 - 90 交併比 |
90 - 0 交併比 |
90 - 90 交併比 |
0.9708 |
2.6525 |
4000 |
0.9827 |
0.3521 |
0.5133 |
0.5141 |
0.6031 |
0.4019 |
0.6068 |
0.4413 |
0.4004 |
0.3106 |
0.2982 |
0.3990 |
1.8598 |
5.3050 |
8000 |
0.6389 |
0.5316 |
0.6914 |
0.6913 |
0.6875 |
0.7105 |
0.5770 |
0.7908 |
0.6004 |
0.4655 |
0.5143 |
0.5463 |
0.4633 |
7.9576 |
12000 |
0.5041 |
0.6369 |
0.7767 |
0.7768 |
0.7883 |
0.7582 |
0.7869 |
0.7735 |
0.6734 |
0.6126 |
0.5897 |
0.6720 |
0.5576 |
10.6101 |
16000 |
0.4221 |
0.6949 |
0.8191 |
0.8194 |
0.8477 |
0.7814 |
0.8269 |
0.8206 |
0.7084 |
0.7075 |
0.6533 |
0.7106 |
0.3003 |
13.2626 |
20000 |
0.3963 |
0.7293 |
0.8413 |
0.8414 |
0.8524 |
0.8809 |
0.8047 |
0.8273 |
0.7565 |
0.6617 |
0.7326 |
0.7666 |
0.3174 |
15.9151 |
24000 |
0.4310 |
0.7511 |
0.8573 |
0.8572 |
0.8538 |
0.8774 |
0.8402 |
0.8577 |
0.7779 |
0.7125 |
0.7412 |
0.7729 |
0.2307 |
18.5676 |
28000 |
0.3326 |
0.8024 |
0.8901 |
0.8902 |
0.9029 |
0.8852 |
0.8832 |
0.8893 |
0.8196 |
0.7874 |
0.7805 |
0.8221 |
0.1946 |
21.2202 |
32000 |
0.2625 |
0.8409 |
0.9134 |
0.9134 |
0.9136 |
0.9197 |
0.8974 |
0.9230 |
0.8533 |
0.8114 |
0.8410 |
0.8581 |
0.1325 |
23.8727 |
36000 |
0.1298 |
0.9185 |
0.9575 |
0.9576 |
0.9647 |
0.9488 |
0.9691 |
0.9474 |
0.9319 |
0.9114 |
0.9158 |
0.9150 |
0.1007 |
26.5252 |
40000 |
0.0752 |
0.9596 |
0.9794 |
0.9794 |
0.9824 |
0.9811 |
0.9748 |
0.9792 |
0.9671 |
0.9557 |
0.9532 |
0.9624 |
0.0336 |
29.1777 |
44000 |
0.2640 |
0.9435 |
0.9709 |
0.9709 |
0.9750 |
0.9668 |
0.9713 |
0.9706 |
0.9473 |
0.9389 |
0.9382 |
0.9497 |
0.0182 |
31.8302 |
48000 |
0.1066 |
0.9680 |
0.9837 |
0.9837 |
0.9882 |
0.9801 |
0.9845 |
0.9821 |
0.9712 |
0.9677 |
0.9627 |
0.9703 |
0.0141 |
34.4828 |
52000 |
0.0716 |
0.9806 |
0.9902 |
0.9902 |
0.9880 |
0.9898 |
0.9893 |
0.9937 |
0.9822 |
0.9789 |
0.9792 |
0.9821 |
0.0117 |
37.1353 |
56000 |
0.0705 |
0.9850 |
0.9925 |
0.9925 |
0.9914 |
0.9920 |
0.9929 |
0.9935 |
0.9861 |
0.9838 |
0.9842 |
0.9859 |
0.0129 |
39.7878 |
60000 |
0.0932 |
0.9833 |
0.9916 |
0.9916 |
0.9897 |
0.9930 |
0.9920 |
0.9916 |
0.9834 |
0.9823 |
0.9827 |
0.9846 |
0.0101 |
42.4403 |
64000 |
0.0302 |
0.9924 |
0.9962 |
0.9962 |
0.9964 |
0.9966 |
0.9972 |
0.9946 |
0.9938 |
0.9917 |
0.9921 |
0.9921 |
0.0071 |
45.0928 |
68000 |
0.0294 |
0.9933 |
0.9966 |
0.9966 |
0.9971 |
0.9961 |
0.9958 |
0.9976 |
0.9946 |
0.9931 |
0.9925 |
0.9930 |
0.0066 |
47.7454 |
72000 |
0.0637 |
0.9912 |
0.9956 |
0.9956 |
0.9959 |
0.9957 |
0.9962 |
0.9946 |
0.9927 |
0.9917 |
0.9895 |
0.9910 |
0.0083 |
50.3979 |
76000 |
0.0494 |
0.9901 |
0.9950 |
0.9950 |
0.9956 |
0.9951 |
0.9946 |
0.9948 |
0.9901 |
0.9899 |
0.9887 |
0.9917 |
0.0049 |
53.0504 |
80000 |
0.0286 |
0.9944 |
0.9972 |
0.9972 |
0.9982 |
0.9967 |
0.9976 |
0.9963 |
0.9956 |
0.9935 |
0.9939 |
0.9946 |
0.0104 |
55.7029 |
84000 |
0.0244 |
0.9953 |
0.9976 |
0.9976 |
0.9984 |
0.9971 |
0.9978 |
0.9971 |
0.9957 |
0.9953 |
0.9942 |
0.9959 |
0.0046 |
58.3554 |
88000 |
0.0298 |
0.9952 |
0.9976 |
0.9976 |
0.9983 |
0.9969 |
0.9980 |
0.9971 |
0.9959 |
0.9950 |
0.9945 |
0.9954 |
框架版本
- Transformers 4.48.3
- Pytorch 2.1.0
- Datasets 3.2.0
- Tokenizers 0.21.0
📄 許可證
該項目使用其他類型的許可證。
屬性 |
詳情 |
庫名稱 |
transformers |
許可證 |
其他 |
基礎模型 |
nvidia/mit - b3 |
標籤 |
視覺、圖像分割、由訓練器生成 |
模型名稱 |
segformer - b0 - finetuned - morphpadver1 - hgo - coord - v3 |