Hugging Face's logo

VBoussot
/
ImpactSynth

Image-to-Image
konfai
medical-imaging
synthetic-ct
image-synthesis
radiotherapy
mri
cbct
totalsynth
Model card Files
xet
 Community
VBoussot commited on
Commit
d382d12
·
verified ·
1 Parent(s): db04a8e

Add IMPACT-Synth training config (losses + optimizer) so the app is fine-tunable

Browse files

Restore the training loss the shipped prediction config lacked: MAE + SAM_Perceptual (SAM2.1) + L1 on the Tanh head vs CT, a masked MAE metric, AdamW + StepLR. Output key Tanh verified against UNetPlusPlus.yml. Enables fine_tune (was a no-op RESUME with outputs_criterions: None).

Files changed (4) hide show
  1. CBCT/Config.yml +62 -1
  2. Finetune/Config.yml +62 -1
  3. MR/Config.yml +62 -1
  4. MR_CBCT/Config.yml +62 -1
CBCT/Config.yml CHANGED
@@ -2,7 +2,68 @@ Trainer:
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
- outputs_criterions: None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  Dataset:
7
  groups_src:
8
  MASK:
 
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
+ outputs_criterions:
6
+ Tanh:
7
+ targets_criterions:
8
+ CT:
9
+ criterions_loader:
10
+ MAE:
11
+ is_loss: true
12
+ schedulers: {Constant: {nb_step: 0, value: 1}}
13
+ group: 0
14
+ start: 0
15
+ stop: None
16
+ accumulation: false
17
+ reduction: mean
18
+ SAM_Perceptual:
19
+ is_loss: true
20
+ schedulers: {Constant: {nb_step: 0, value: 1}}
21
+ group: 0
22
+ start: 0
23
+ stop: None
24
+ accumulation: false
25
+ train: true
26
+ model_name: SAM2.1_Small.pt
27
+ weights: [1, 1, 1]
28
+ torch:nn:L1Loss:
29
+ size_average: None
30
+ reduce: None
31
+ reduction: mean
32
+ schedulers: {Constant: {nb_step: 0, value: 1}}
33
+ is_loss: true
34
+ group: 0
35
+ start: 0
36
+ stop: None
37
+ accumulation: false
38
+ CT;MASK:
39
+ criterions_loader:
40
+ MAE:
41
+ schedulers: {Constant: {nb_step: 0, value: 1}}
42
+ is_loss: false
43
+ group: 0
44
+ start: 0
45
+ stop: None
46
+ accumulation: false
47
+ reduction: mean
48
+ optimizer:
49
+ name: AdamW
50
+ lr: 0.001
51
+ betas: [0.9, 0.999]
52
+ eps: 1e-08
53
+ weight_decay: 0.001
54
+ amsgrad: false
55
+ maximize: false
56
+ foreach: None
57
+ capturable: false
58
+ differentiable: false
59
+ fused: None
60
+ schedulers:
61
+ StepLR:
62
+ step_size: 10
63
+ gamma: 0.75
64
+ last_epoch: -1
65
+ verbose: deprecated
66
+ nb_step: 0
67
  Dataset:
68
  groups_src:
69
  MASK:
Finetune/Config.yml CHANGED
@@ -2,7 +2,68 @@ Trainer:
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
- outputs_criterions: None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  Dataset:
7
  groups_src:
8
  MASK:
 
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
+ outputs_criterions:
6
+ Tanh:
7
+ targets_criterions:
8
+ CT:
9
+ criterions_loader:
10
+ MAE:
11
+ is_loss: true
12
+ schedulers: {Constant: {nb_step: 0, value: 1}}
13
+ group: 0
14
+ start: 0
15
+ stop: None
16
+ accumulation: false
17
+ reduction: mean
18
+ SAM_Perceptual:
19
+ is_loss: true
20
+ schedulers: {Constant: {nb_step: 0, value: 1}}
21
+ group: 0
22
+ start: 0
23
+ stop: None
24
+ accumulation: false
25
+ train: true
26
+ model_name: SAM2.1_Small.pt
27
+ weights: [1, 1, 1]
28
+ torch:nn:L1Loss:
29
+ size_average: None
30
+ reduce: None
31
+ reduction: mean
32
+ schedulers: {Constant: {nb_step: 0, value: 1}}
33
+ is_loss: true
34
+ group: 0
35
+ start: 0
36
+ stop: None
37
+ accumulation: false
38
+ CT;MASK:
39
+ criterions_loader:
40
+ MAE:
41
+ schedulers: {Constant: {nb_step: 0, value: 1}}
42
+ is_loss: false
43
+ group: 0
44
+ start: 0
45
+ stop: None
46
+ accumulation: false
47
+ reduction: mean
48
+ optimizer:
49
+ name: AdamW
50
+ lr: 0.001
51
+ betas: [0.9, 0.999]
52
+ eps: 1e-08
53
+ weight_decay: 0.001
54
+ amsgrad: false
55
+ maximize: false
56
+ foreach: None
57
+ capturable: false
58
+ differentiable: false
59
+ fused: None
60
+ schedulers:
61
+ StepLR:
62
+ step_size: 10
63
+ gamma: 0.75
64
+ last_epoch: -1
65
+ verbose: deprecated
66
+ nb_step: 0
67
  Dataset:
68
  groups_src:
69
  MASK:
MR/Config.yml CHANGED
@@ -2,7 +2,68 @@ Trainer:
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
- outputs_criterions: None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  Dataset:
7
  groups_src:
8
  MASK:
 
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
+ outputs_criterions:
6
+ Tanh:
7
+ targets_criterions:
8
+ CT:
9
+ criterions_loader:
10
+ MAE:
11
+ is_loss: true
12
+ schedulers: {Constant: {nb_step: 0, value: 1}}
13
+ group: 0
14
+ start: 0
15
+ stop: None
16
+ accumulation: false
17
+ reduction: mean
18
+ SAM_Perceptual:
19
+ is_loss: true
20
+ schedulers: {Constant: {nb_step: 0, value: 1}}
21
+ group: 0
22
+ start: 0
23
+ stop: None
24
+ accumulation: false
25
+ train: true
26
+ model_name: SAM2.1_Small.pt
27
+ weights: [1, 1, 1]
28
+ torch:nn:L1Loss:
29
+ size_average: None
30
+ reduce: None
31
+ reduction: mean
32
+ schedulers: {Constant: {nb_step: 0, value: 1}}
33
+ is_loss: true
34
+ group: 0
35
+ start: 0
36
+ stop: None
37
+ accumulation: false
38
+ CT;MASK:
39
+ criterions_loader:
40
+ MAE:
41
+ schedulers: {Constant: {nb_step: 0, value: 1}}
42
+ is_loss: false
43
+ group: 0
44
+ start: 0
45
+ stop: None
46
+ accumulation: false
47
+ reduction: mean
48
+ optimizer:
49
+ name: AdamW
50
+ lr: 0.001
51
+ betas: [0.9, 0.999]
52
+ eps: 1e-08
53
+ weight_decay: 0.001
54
+ amsgrad: false
55
+ maximize: false
56
+ foreach: None
57
+ capturable: false
58
+ differentiable: false
59
+ fused: None
60
+ schedulers:
61
+ StepLR:
62
+ step_size: 10
63
+ gamma: 0.75
64
+ last_epoch: -1
65
+ verbose: deprecated
66
+ nb_step: 0
67
  Dataset:
68
  groups_src:
69
  MASK:
MR_CBCT/Config.yml CHANGED
@@ -2,7 +2,68 @@ Trainer:
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
- outputs_criterions: None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  Dataset:
7
  groups_src:
8
  MASK:
 
2
  Model:
3
  classpath: UNetPlusPlus.yml
4
  UNetPlusPlus:
5
+ outputs_criterions:
6
+ Tanh:
7
+ targets_criterions:
8
+ CT:
9
+ criterions_loader:
10
+ MAE:
11
+ is_loss: true
12
+ schedulers: {Constant: {nb_step: 0, value: 1}}
13
+ group: 0
14
+ start: 0
15
+ stop: None
16
+ accumulation: false
17
+ reduction: mean
18
+ SAM_Perceptual:
19
+ is_loss: true
20
+ schedulers: {Constant: {nb_step: 0, value: 1}}
21
+ group: 0
22
+ start: 0
23
+ stop: None
24
+ accumulation: false
25
+ train: true
26
+ model_name: SAM2.1_Small.pt
27
+ weights: [1, 1, 1]
28
+ torch:nn:L1Loss:
29
+ size_average: None
30
+ reduce: None
31
+ reduction: mean
32
+ schedulers: {Constant: {nb_step: 0, value: 1}}
33
+ is_loss: true
34
+ group: 0
35
+ start: 0
36
+ stop: None
37
+ accumulation: false
38
+ CT;MASK:
39
+ criterions_loader:
40
+ MAE:
41
+ schedulers: {Constant: {nb_step: 0, value: 1}}
42
+ is_loss: false
43
+ group: 0
44
+ start: 0
45
+ stop: None
46
+ accumulation: false
47
+ reduction: mean
48
+ optimizer:
49
+ name: AdamW
50
+ lr: 0.001
51
+ betas: [0.9, 0.999]
52
+ eps: 1e-08
53
+ weight_decay: 0.001
54
+ amsgrad: false
55
+ maximize: false
56
+ foreach: None
57
+ capturable: false
58
+ differentiable: false
59
+ fused: None
60
+ schedulers:
61
+ StepLR:
62
+ step_size: 10
63
+ gamma: 0.75
64
+ last_epoch: -1
65
+ verbose: deprecated
66
+ nb_step: 0
67
  Dataset:
68
  groups_src:
69
  MASK: