pytorch_connectomics/tutorials/vesicle_xm.yaml at master · PytorchConnectomics/pytorch_connectomics · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
experiment_name: vesicle_xm_mednext_bcd
description: Vesicle segmentation with MedNeXt and multi-task learning (Binary + Contour + DT)

_base_:
- ../connectomics/config/all_profiles.yaml

default:
  pipeline_profile: bcd
  system:
    profile: all-gpu-cpu
  model:
    arch:
      profile: mednext_s
    input_size: [32, 128, 128]
    output_size: [32, 128, 128]
  data:
    dataloader:
      profile: cached
      batch_size: 4
      patch_size: [32, 128, 128]
    image_transform:
      normalize: "0-1"
    augmentation:
      profile: aug_light
  inference:
    test_time_augmentation:
      enabled: true
      flip_axes: [1, 2]
      rotation90_axes: [[1, 2]]
    sliding_window:
      window_size: [32, 128, 128]
    save_prediction:
      enabled: true
  decoding:
    steps:
    - name: decode_instance_binary_contour_distance
      kwargs:
        binary_threshold: [0.76, 0.5]
        contour_threshold: [0.6, 1]
        distance_threshold: [0.7, -0.2]
        min_instance_size: 100
        min_seed_size: 20
    postprocessing:
      enabled: true
      instance_cc3d:
        connectivity: 6        # 6-connected (face only) for conservative splitting
        min_size: 100           # remove components smaller than 100 voxels

  evaluation:
    enabled: true
    metrics:
    - adapted_rand

train:
  data:
    train:
      image: ["datasets/bouton-lv/train/vol0_im_bouton.h5", "datasets/hydra-lv/vol*_im.h5"]
      label: ["datasets/bouton-lv/train/vol0_lv_bouton.h5", "datasets/hydra-lv/vol*_vesicle_ins.h5"]
      mask: ["datasets/bouton-lv/train/vol0_mask_bouton.h5", "datasets/hydra-lv/vol*_mask.h5"]
      resolution: [30, 8, 8]
    # make sure samples have enough foreground for masked sampling
    dataloader:
      cached_sampling_foreground_threshold: 0.05
      cached_sampling_sample_nonzero_mask: true
  optimization:
    profile: warmup_cosine_lr
    n_steps_per_epoch: 1000
    max_epochs: 500

  monitor:
    logging:
      scalar:
        loss_every_n_steps: 50
      images:
        log_every_n_epochs: 10
        max_images: 8
        num_slices: 2
    checkpoint:
      save_top_k: 3
      monitor: train_loss_total_epoch

test:
  data:
    test:
      path: "/projects/weilab/dataset/xiaomeng/vesicle/"
      # debugging
      image: "im/Bouton_228_EM.tif"
      mask: "mask/Bouton_228_em_mask.tif"
      #image: "im/Bouton_*_EM.tif"
      #mask: "mask/Bouton_*_em_mask.tif"
      resolution: [30, 8, 8]
    # fix: the exported mask is 2x downsampled, so we need to upsample it to the original size
    mask_transform:
      resize: [1, 2, 2]
      align_to_image: true
      binarize: true

# ============================================================================
# Parameter tuning for BCD watershed thresholds (--mode tune)
# ============================================================================
tune:
  profile: tune_bcd
  n_trials: 50
  study_name: vesicle_xm_bcd_tuning
  data:
    val:
      image: ["datasets/bouton-lv/train/vol0_im_bouton.h5", "datasets/hydra-lv/vol*_im.h5"]
      label: ["datasets/bouton-lv/train/vol0_lv_bouton.h5", "datasets/hydra-lv/vol*_vesicle_ins.h5"]
      mask: ["datasets/bouton-lv/train/vol0_mask_bouton.h5", "datasets/hydra-lv/vol*_mask.h5"]
  # Override profile defaults for vesicle-specific search ranges
  parameter_space:
    decoding:
      defaults:
        binary_threshold: [0.9, 0.85]
        contour_threshold: [0.8, 1]
        distance_threshold: [0.5, -0.5]
        min_instance_size: 30
        min_seed_size: 5
      parameters:
        binary_threshold_seed:
          range: [0.7, 0.99]
          step: 0.01
        binary_threshold_fg:
          range: [0.5, 0.95]
          step: 0.05
        contour_threshold_low:
          range: [0.3, 0.95]
          step: 0.05
        contour_threshold_high:
          range: [0.8, 1]
          step: 0.1
        distance_threshold_seed:
          range: [0.0, 1.0]
          step: 0.1
        distance_threshold_fg:
          range: [-1.0, 0.5]
          step: 0.1
        min_instance_size:
          range: [10, 200]
          step: 10
        min_seed_size:
          range: [1, 30]
          step: 1