pytorch_connectomics/tutorials/waterz_decoding.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
experiment_name: waterz_decode_only
description: >
  Decode-only: load a pre-computed affinity HDF5 and run waterz decoding.
  No model, no inference, no TTA.

  Usage:
    python scripts/main.py --config tutorials/waterz_decoding.yaml --mode test \
      decoding.input_prediction_path=path/to/affinity.h5 \
      test.data.test.label=path/to/gt.h5

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

default:
  inference: {}
  decoding:
    steps:
    - template: decoding_waterz
      kwargs:
        use_aff_uint8: true
        use_seg_uint32: true
        thresholds: 0.4
        merge_function: affmean
        #merge_function: aff85_his256
        aff_threshold: [0, 1]
        channel_order: xyz
        boundary_threshold: 0.3
        dust_merge: true
        dust_merge_size: 400
        dust_merge_affinity: 0.3
        dust_remove_size: 200

    input_prediction_path: "/projects/weilab/dataset/liconn/mansour/DL288B_251222S_cond5_40x_12tiles_round1_fused_488_crop512x1024x1024_ffn_sharp_tta_x8_prediction_uint8.h5"
test:
  data:
    test:
      resolution: [25, 18, 18]