test_unet_sem.yaml

 1mode: test
 2# contents to define a model
 3model:
 4## model architecture, SeM indicates Segmentation Model
 5  name: SeM
 6  ## loss function of architecture. For SeM, you need to specify the segmentation loss.
 7  segmentation_losses:
 8    - name: Dice
 9      weight: 1.0
10    - name: BCEL
11      weight: 1.0
12  ## define encoder and decoder
13  encoder:
14    ### encoder name
15    name: UNet
16    ### input image size (after augmented)
17    image_size: [320, 256]
18    ### number of input image channels
19    in_channel: 3
20    ### number of output channels, for segmentation, it should be the number of categories
21    out_channel: 1
22    ### number of channel for image patches
23    patch_channel: 32
24    ### size of image patch, for 2D image, 'patch size = 2' indicates a 2*2 image patch.
25    patch_size: 2
26    ### number of channel for each layer in down-sample layers.
27    ### The length of list is the number of down-sample layers
28    down_channels: [64, 128, 256]
29    ### number of channel for each layer in middle layers.
30    ### The length of list is the number of middle layers
31    middle_channels: [512, 512]
32    ### building block
33    building_block: conv
34    ### normalization
35    normalization: batch
36# data loader for test
37loader:
38  ## define a dataset
39  dataset:
40    name: ImgSegDataset
41    data_dir: images
42    label_dir: masks
43    data_suffix: jpg
44  ## merge listed datasets into one
45  data_path_list:
46    - path/to/datasets/CVC-ClinicDB/fold5/
47  ### other parameters related to dataloader
48  ### refer to https://pytorch.org/docs/stable/data.html#torch.utils.data.DataLoader
49  batch_size: 16
50  num_workers: 8
51  shuffle: true
52  ## augmentations performed on each sample from the dataset
53  data_transforms:
54    - name: Resize
55      size: [320, 256]
56    - name: ToTensor
57# evaluate test dataset
58evaluation_metrics:
59  seg:
60    - name: Dice
61    - name: ACC
62    - name: mIoU
63
64model_path: path/to/checkpoint/CVC-ClinicDB/UNet/ckp_last.pth
65# directory for saving results
66seg_dir: path/to/results/seg/CVC-ClinicDB/UNet
67# save transformed target
68save_target: true
69# save transformed input
70save_input: true
71# save colorized segmentation results
72save_colorized: true