要做的 from torchinfo import summary刪除 增加訓練教學 測試結合image
2021/12/24增加內容
更改test.py 解決model無法正確載入問題
modify test.py solve model can't load correctlt
增加fcn8, fcn32, UNet, UNet++訓練300 epochs之權重
add fcn8, fcn32, UNet, UNet++ weight with train 300 epochs
增加fcn8, fcn32, UNet, UNet++之預測圖片
add fcn8, fcn32, UNet, UNet++ predict images
增加fcn8, fcn32, UNet, UNet++之預測圖片使用eval.py所產生之txt檔案
add fcn8, fcn32, UNet, UNet++ performance txt file produce by eval.py using predict images
Semantic-Segmentation-Multiclass_FCN_Unet_ResUnet(Multiclass)
run train.py 執行train.py即可訓練
run test.py 執行test.py即可使用所訓練的pt檔預測輸出，輸出目錄為dataset/result，檔名會新增_predict以辨識
run eval.py 執行eval.py即可評估dataset/result內的輸出圖像跟ground truth圖像，將會在目錄輸出Model_Predict.txt，內含Average Pixel Accuracy | classPixelAccuracy | Average Mean Accuracy | Average Mean IU
Python 3.6
Pytorch 1.8
Cuda 10.2
input images are resized to [256 ,256]
Training was performed using Stochastic Gradient Descent (SGD) with learning rate 0.001
輸入圖像被縮放至[256 ,256] 使用SGD，學習率0.001
The dataset is divided into three parts: 160 images for the training set and 40 images for the validation set
The dataset is to segment 4 classes such as circle,square,triangle,star
資料集被分為160張用來訓練，40張用來驗證，並且有4個類[circle,square,triangle,star]
Their corresponding grayscale values are
| Class | Grayscale |
|---|---|
| circle | 100 |
| square | 250 |
| triangle | 255 |
| star | 200 |
| Model | Average Pixel Accuracy | Average Mean Accuracy | Circle Accuracy | Square Accuracy | Triangle Accuracy | Star Accuracy | Background Accuracy |
|---|---|---|---|---|---|---|---|
| FCN8S | 0.941 | 0.413 | 0.120 | 0.438 | 0.186 | 0.323 | 0.998 |
| FCN32S | 0.934 | 0.331 | 0.176 | 0.333 | 0.036 | 0.112 | 0.998 |
| UNet | 0.980 | 0.827 | 0.791 | 0.838 | 0.733 | 0.775 | 0.996 |
| UNet_Plus | 0.979 | 0.802 | 0.770 | 0.851 | 0.694 | 0.696 | 0.997 |
| ResUNet | 0.982 | 0.827 | 0.797 | 0.852 | 0.687 | 0.800 | 0.998 |
Ground Truth FCN8S FCN32S
UNet UNet Plus ResUNet