akramoo commited on
Commit
9baf699
·
verified ·
1 Parent(s): fa4163f

Upload folder using huggingface_hub

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. README.md +166 -0
  2. virchow2_abmil_focal_maskgate_5seed/seed0/best_model.safetensors +3 -0
  3. virchow2_abmil_focal_maskgate_5seed/seed0/config.json +25 -0
  4. virchow2_abmil_focal_maskgate_5seed/seed0/release_info.json +11 -0
  5. virchow2_abmil_focal_maskgate_5seed/seed1/best_model.safetensors +3 -0
  6. virchow2_abmil_focal_maskgate_5seed/seed1/config.json +25 -0
  7. virchow2_abmil_focal_maskgate_5seed/seed1/release_info.json +11 -0
  8. virchow2_abmil_focal_maskgate_5seed/seed2/best_model.safetensors +3 -0
  9. virchow2_abmil_focal_maskgate_5seed/seed2/config.json +25 -0
  10. virchow2_abmil_focal_maskgate_5seed/seed2/release_info.json +11 -0
  11. virchow2_abmil_focal_maskgate_5seed/seed3/best_model.safetensors +3 -0
  12. virchow2_abmil_focal_maskgate_5seed/seed3/config.json +25 -0
  13. virchow2_abmil_focal_maskgate_5seed/seed3/release_info.json +11 -0
  14. virchow2_abmil_focal_maskgate_5seed/seed4/best_model.safetensors +3 -0
  15. virchow2_abmil_focal_maskgate_5seed/seed4/config.json +26 -0
  16. virchow2_abmil_focal_maskgate_5seed/seed4/release_info.json +11 -0
  17. virchow2_transmil_coral_fullgate_5fold/fold_1/best_model.safetensors +3 -0
  18. virchow2_transmil_coral_fullgate_5fold/fold_1/config.json +27 -0
  19. virchow2_transmil_coral_fullgate_5fold/fold_1/release_info.json +11 -0
  20. virchow2_transmil_coral_fullgate_5fold/fold_2/best_model.safetensors +3 -0
  21. virchow2_transmil_coral_fullgate_5fold/fold_2/config.json +27 -0
  22. virchow2_transmil_coral_fullgate_5fold/fold_2/release_info.json +11 -0
  23. virchow2_transmil_coral_fullgate_5fold/fold_3/best_model.safetensors +3 -0
  24. virchow2_transmil_coral_fullgate_5fold/fold_3/config.json +27 -0
  25. virchow2_transmil_coral_fullgate_5fold/fold_3/release_info.json +11 -0
  26. virchow2_transmil_coral_fullgate_5fold/fold_4/best_model.safetensors +3 -0
  27. virchow2_transmil_coral_fullgate_5fold/fold_4/config.json +27 -0
  28. virchow2_transmil_coral_fullgate_5fold/fold_4/release_info.json +11 -0
  29. virchow2_transmil_coral_fullgate_5fold/fold_5/best_model.safetensors +3 -0
  30. virchow2_transmil_coral_fullgate_5fold/fold_5/config.json +27 -0
  31. virchow2_transmil_coral_fullgate_5fold/fold_5/release_info.json +11 -0
  32. virchow2_transmil_coral_maskgate_5fold/fold_1/best_model.safetensors +3 -0
  33. virchow2_transmil_coral_maskgate_5fold/fold_1/config.json +27 -0
  34. virchow2_transmil_coral_maskgate_5fold/fold_1/release_info.json +11 -0
  35. virchow2_transmil_coral_maskgate_5fold/fold_2/best_model.safetensors +3 -0
  36. virchow2_transmil_coral_maskgate_5fold/fold_2/config.json +27 -0
  37. virchow2_transmil_coral_maskgate_5fold/fold_2/release_info.json +11 -0
  38. virchow2_transmil_coral_maskgate_5fold/fold_3/best_model.safetensors +3 -0
  39. virchow2_transmil_coral_maskgate_5fold/fold_3/config.json +27 -0
  40. virchow2_transmil_coral_maskgate_5fold/fold_3/release_info.json +11 -0
  41. virchow2_transmil_coral_maskgate_5fold/fold_4/best_model.safetensors +3 -0
  42. virchow2_transmil_coral_maskgate_5fold/fold_4/config.json +27 -0
  43. virchow2_transmil_coral_maskgate_5fold/fold_4/release_info.json +11 -0
  44. virchow2_transmil_coral_maskgate_5fold/fold_5/best_model.safetensors +3 -0
  45. virchow2_transmil_coral_maskgate_5fold/fold_5/config.json +27 -0
  46. virchow2_transmil_coral_maskgate_5fold/fold_5/release_info.json +11 -0
  47. virchow2_transmil_focal_fullgate_5seed/seed0/best_model.safetensors +3 -0
  48. virchow2_transmil_focal_fullgate_5seed/seed0/config.json +26 -0
  49. virchow2_transmil_focal_fullgate_5seed/seed0/release_info.json +11 -0
  50. virchow2_transmil_focal_fullgate_5seed/seed1/best_model.safetensors +3 -0
README.md CHANGED
@@ -1,3 +1,169 @@
1
  ---
2
  license: cc-by-nc-sa-4.0
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-nc-sa-4.0
3
+ tags:
4
+ - pathology
5
+ - computational-pathology
6
+ - histopathology
7
+ - multiple-instance-learning
8
+ - whole-slide-images
9
+ - prostate-cancer
10
+ - medical-imaging
11
+ - pytorch
12
  ---
13
+
14
+ # Artifact-aware MIL heads for prostate biopsy ISUP grading
15
+
16
+ Trained multiple instance learning (MIL) heads from the paper *Aggregation,
17
+ Ordinal Training, and Artifact Gating in Multiple Instance Learning for
18
+ Prostate Biopsy Grading*. Each model maps a bag of frozen Virchow2 patch
19
+ embeddings from one prostate biopsy whole-slide image to an ISUP grade group
20
+ (0–5).
21
+
22
+ Code, data split and result files: <https://github.com/PathOS-Code/artifact-aware-mil-prostate>
23
+
24
+ > **Research use only. Not for clinical use.** These models have not been
25
+ > validated for diagnosis or treatment decisions and are not a medical device.
26
+
27
+ ## Contents
28
+
29
+ 30 models, the members of every ensemble reported in the paper. Each folder
30
+ holds `best_model.safetensors` (weights only), the run's recorded
31
+ `config.json`, and `release_info.json` (best epoch, validation QWK, SHA-256).
32
+
33
+ | Folder | Members | Aggregator | Loss | Gate | Protocol |
34
+ | --- | --- | --- | --- | --- | --- |
35
+ | `virchow2_transmil_focal_nogate_5seed/` | `seed0`–`seed4` | TransMIL-style, 2 layers | focal | none | fixed split, 5 seeds |
36
+ | `virchow2_transmil_focal_maskgate_5seed/` | `seed0`–`seed4` | TransMIL-style, 2 layers | focal | mask-only | fixed split, 5 seeds |
37
+ | `virchow2_transmil_focal_fullgate_5seed/` | `seed0`–`seed4` | TransMIL-style, 2 layers | focal | full robust | fixed split, 5 seeds |
38
+ | `virchow2_abmil_focal_maskgate_5seed/` | `seed0`–`seed4` | ABMIL | focal | mask-only | fixed split, 5 seeds |
39
+ | `virchow2_transmil_coral_maskgate_5fold/` | `fold_1`–`fold_5` | TransMIL-style, 2 layers | CORAL | mask-only | 5-fold CV |
40
+ | `virchow2_transmil_coral_fullgate_5fold/` | `fold_1`–`fold_5` | TransMIL-style, 2 layers | CORAL | full robust | 5-fold CV |
41
+
42
+ The mask-only and full robust gate variants implement the same model: flagged
43
+ tokens are excluded from attention in both, so the attenuation factor α does
44
+ not change the output. Their differences are run-to-run (replicate)
45
+ differences. α was not logged for any run; the evaluation code uses the
46
+ default α = 1e-4.
47
+
48
+ Total size about 4.1 GB (TransMIL-style members 163 MB each, ABMIL members
49
+ 6 MB each). The weights were exported from the training checkpoints with
50
+ `scripts/training/export_weights.py` in the code repository; for every model
51
+ the exported weights give identical logits to the original checkpoint.
52
+
53
+ ## Input
54
+
55
+ The models do **not** contain an image encoder. Each slide must be converted
56
+ to a bag saved as a `.pt` file with:
57
+
58
+ | Key | Shape | Content |
59
+ | --- | --- | --- |
60
+ | `embeddings` | `[N, 1280]` | Virchow2 CLS token of each tissue patch |
61
+ | `spatial_variances` | `[N]` | variance of the Laplacian of the grayscale patch, used by the gate |
62
+ | `coords` | `[N, 2]` | patch level-0 coordinates (used only for heatmaps) |
63
+
64
+ Patches are 256 × 256 pixels at level 0 (nominal 20×), Macenko stain
65
+ normalization with a fixed reference, resized to 224 × 224 for Virchow2. The
66
+ code repository's `scripts/preprocessing/preprocess.py` and
67
+ `scripts/feature_extraction/extract_features.py` produce these bags.
68
+
69
+ **Virchow2 weights are not included.** Request access to
70
+ [`paige-ai/Virchow2`](https://huggingface.co/paige-ai/Virchow2) and use it
71
+ under its own license terms.
72
+
73
+ ## Output
74
+
75
+ - Focal-loss models: 6 logits (ISUP 0–5); class probabilities by softmax.
76
+ - CORAL models: 5 ordinal threshold logits; class probabilities by
77
+ differencing the cumulative sigmoids, clipping at 0 and renormalizing.
78
+
79
+ Ensembles average the class probabilities of their members.
80
+
81
+ ## Usage
82
+
83
+ ```bash
84
+ git clone https://github.com/PathOS-Code/artifact-aware-mil-prostate.git
85
+ cd artifact-aware-mil-prostate
86
+ uv sync
87
+ hf download PathOS-Code/artifact-aware-mil-prostate --local-dir weights
88
+
89
+ uv run python scripts/evaluation/ensemble_eval.py \
90
+ --exp-dirs weights/virchow2_transmil_focal_fullgate_5seed/seed{0,1,2,3,4} \
91
+ --embeddings-dir outputs/embeddings_virchow2 \
92
+ --labels-csv data/panda_labels.csv \
93
+ --split-csv splits/panda_fixed_split.csv \
94
+ --out outputs/ensembles/panda_focal_fullgate_5seed.json
95
+ ```
96
+
97
+ For an external cohort, pass its embeddings and labels CSV with
98
+ `--val-ratio 0 --test-ratio 0` to evaluate every slide. See `REPRODUCE.md` in
99
+ the code repository for all experiments.
100
+
101
+ ## Results
102
+
103
+ From the paper and the result files in the code repository (`results/`).
104
+ QWK = quadratic weighted kappa.
105
+
106
+ PANDA held-out test split (1,061 slides; 10,614 slides overall from Radboud
107
+ and Karolinska):
108
+
109
+ | Ensemble | Test QWK |
110
+ | --- | --- |
111
+ | TransMIL-style, focal, no gate, 5 seeds | 0.9063 |
112
+ | TransMIL-style, focal, mask-only, 5 seeds | 0.9104 |
113
+ | TransMIL-style, focal, full robust gate, 5 seeds | 0.9134 (95% CI 0.8856–0.9368) |
114
+ | ABMIL, focal, mask-only, 5 seeds | 0.8919 |
115
+
116
+ PANDA five-fold cross-validation (out-of-fold QWK, optimistic): CORAL
117
+ mask-only 0.9226, CORAL full robust gate 0.9227.
118
+
119
+ External cohorts, direct transfer without fine-tuning:
120
+
121
+ | Ensemble | SICAPv2 (155 biopsies) | CrowdGleason (232 slide identifiers) |
122
+ | --- | --- | --- |
123
+ | TransMIL-style, focal, no gate, 5 seeds | 0.8664 | 0.8076 |
124
+ | TransMIL-style, focal, mask-only, 5 seeds | 0.8633 | 0.8028 |
125
+ | TransMIL-style, focal, full robust gate, 5 seeds | 0.8482 | 0.8076 |
126
+ | TransMIL-style, CORAL, mask-only, 5 folds | 0.8760 | 0.8109 |
127
+ | TransMIL-style, CORAL, full robust gate, 5 folds | 0.8772 (95% CI 0.803–0.933) | 0.8080 (95% CI 0.733–0.871) |
128
+
129
+ Caveats, as stated in the paper:
130
+
131
+ - The PANDA test ensemble was the best of three ensembles, selected post hoc
132
+ after exploratory runs that included the test slides.
133
+ - None of the paired gated-versus-ungated QWK intervals excluded zero; no
134
+ statistically detectable gating effect was found, and gate firing was not
135
+ measured.
136
+ - All intervals treat slides as independent and ignore patient clustering.
137
+ - CrowdGleason slide labels are surrogates derived from patch labels.
138
+ - Aggregator and ordinal-loss contrasts are confounded and descriptive.
139
+ - Training is not bit-wise reproducible; retrained models differ slightly.
140
+
141
+ ## Limitations
142
+
143
+ - Developed on PANDA only (two institutions, routine-practice labels);
144
+ external evaluation covers two small public cohorts.
145
+ - Performance under other scanners, stains, tissue preparation or patient
146
+ populations is unknown.
147
+ - Grade groups are predicted per slide; no localization of Gleason patterns
148
+ is provided by these heads.
149
+ - Depends on Virchow2 embeddings; other encoders will not work with these
150
+ weights.
151
+
152
+ ## License
153
+
154
+ The weights are released under
155
+ [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/)
156
+ because they were trained on PANDA (CC BY-NC-SA 4.0): noncommercial use,
157
+ attribution required, derivatives under the same license. The code in the
158
+ GitHub repository is under the PolyForm Noncommercial License 1.0.0.
159
+
160
+ ## Citation
161
+
162
+ ```bibtex
163
+ @article{TODO,
164
+ title = {Aggregation, Ordinal Training, and Artifact Gating in Multiple Instance Learning for Prostate Biopsy Grading},
165
+ author = {TODO},
166
+ journal = {TODO},
167
+ year = {TODO}
168
+ }
169
+ ```
virchow2_abmil_focal_maskgate_5seed/seed0/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6232708782c90705c4f5a7fe737fd434cf0310cb2f1eeb2eed1864caf256fe64
3
+ size 5920172
virchow2_abmil_focal_maskgate_5seed/seed0/config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_abmil_gated_seed0",
3
+ "backbone": "virchow2 (1280d)",
4
+ "gating": true,
5
+ "gating_mode": "v2",
6
+ "dropout": 0.25,
7
+ "patch_dropout": 0.2,
8
+ "early_stopping_patience": 0,
9
+ "cosine_lr": false,
10
+ "epochs": 50,
11
+ "learning_rate": 0.0001,
12
+ "weight_decay": 0.01,
13
+ "focal_gamma": 2.0,
14
+ "accumulation_steps": 8,
15
+ "transmil_layers": "N/A",
16
+ "variance_threshold": 122.0,
17
+ "z_score_factor": 2.0,
18
+ "train_slides": 7961,
19
+ "val_slides": 1592,
20
+ "test_slides": 1061,
21
+ "parameters": 1479815,
22
+ "device": "NVIDIA GeForce RTX 3070",
23
+ "init_seed": 0,
24
+ "split_seed": 42
25
+ }
virchow2_abmil_focal_maskgate_5seed/seed0/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_abmil_focal_maskgate_5seed/seed0",
3
+ "best_epoch": 27,
4
+ "best_val_qwk": 0.9077715569794771,
5
+ "n_parameters": 1479815,
6
+ "weights_sha256": "6232708782c90705c4f5a7fe737fd434cf0310cb2f1eeb2eed1864caf256fe64",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_abmil_focal_maskgate_5seed/seed1/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68740082c74008a4cb131dae8dda8b95223a19a45f74ed1d9bdeed746b4b49e9
3
+ size 5920172
virchow2_abmil_focal_maskgate_5seed/seed1/config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_abmil_gated_seed1",
3
+ "backbone": "virchow2 (1280d)",
4
+ "gating": true,
5
+ "gating_mode": "v2",
6
+ "dropout": 0.25,
7
+ "patch_dropout": 0.2,
8
+ "early_stopping_patience": 0,
9
+ "cosine_lr": false,
10
+ "epochs": 50,
11
+ "learning_rate": 0.0001,
12
+ "weight_decay": 0.01,
13
+ "focal_gamma": 2.0,
14
+ "accumulation_steps": 8,
15
+ "transmil_layers": "N/A",
16
+ "variance_threshold": 122.0,
17
+ "z_score_factor": 2.0,
18
+ "train_slides": 7961,
19
+ "val_slides": 1592,
20
+ "test_slides": 1061,
21
+ "parameters": 1479815,
22
+ "device": "NVIDIA GeForce RTX 3070",
23
+ "init_seed": 1,
24
+ "split_seed": 42
25
+ }
virchow2_abmil_focal_maskgate_5seed/seed1/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_abmil_focal_maskgate_5seed/seed1",
3
+ "best_epoch": 43,
4
+ "best_val_qwk": 0.9106901289987646,
5
+ "n_parameters": 1479815,
6
+ "weights_sha256": "68740082c74008a4cb131dae8dda8b95223a19a45f74ed1d9bdeed746b4b49e9",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_abmil_focal_maskgate_5seed/seed2/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5076b069fbf1e05ba33cff9e6f7245d944d3265b9883906891cfb6c8637e3c32
3
+ size 5920172
virchow2_abmil_focal_maskgate_5seed/seed2/config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_abmil_gated_seed2",
3
+ "backbone": "virchow2 (1280d)",
4
+ "gating": true,
5
+ "gating_mode": "v2",
6
+ "dropout": 0.25,
7
+ "patch_dropout": 0.2,
8
+ "early_stopping_patience": 0,
9
+ "cosine_lr": false,
10
+ "epochs": 50,
11
+ "learning_rate": 0.0001,
12
+ "weight_decay": 0.01,
13
+ "focal_gamma": 2.0,
14
+ "accumulation_steps": 8,
15
+ "transmil_layers": "N/A",
16
+ "variance_threshold": 122.0,
17
+ "z_score_factor": 2.0,
18
+ "train_slides": 7961,
19
+ "val_slides": 1592,
20
+ "test_slides": 1061,
21
+ "parameters": 1479815,
22
+ "device": "NVIDIA GeForce RTX 3070",
23
+ "init_seed": 2,
24
+ "split_seed": 42
25
+ }
virchow2_abmil_focal_maskgate_5seed/seed2/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_abmil_focal_maskgate_5seed/seed2",
3
+ "best_epoch": 32,
4
+ "best_val_qwk": 0.9117707732552237,
5
+ "n_parameters": 1479815,
6
+ "weights_sha256": "5076b069fbf1e05ba33cff9e6f7245d944d3265b9883906891cfb6c8637e3c32",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_abmil_focal_maskgate_5seed/seed3/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd95421639858940513c054f5116a233c29ee449357c5db9fe119ca14bb2ef7c
3
+ size 5920172
virchow2_abmil_focal_maskgate_5seed/seed3/config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_abmil_gated_seed3",
3
+ "backbone": "virchow2 (1280d)",
4
+ "gating": true,
5
+ "gating_mode": "v2",
6
+ "dropout": 0.25,
7
+ "patch_dropout": 0.2,
8
+ "early_stopping_patience": 0,
9
+ "cosine_lr": false,
10
+ "epochs": 50,
11
+ "learning_rate": 0.0001,
12
+ "weight_decay": 0.01,
13
+ "focal_gamma": 2.0,
14
+ "accumulation_steps": 8,
15
+ "transmil_layers": "N/A",
16
+ "variance_threshold": 122.0,
17
+ "z_score_factor": 2.0,
18
+ "train_slides": 7961,
19
+ "val_slides": 1592,
20
+ "test_slides": 1061,
21
+ "parameters": 1479815,
22
+ "device": "NVIDIA GeForce RTX 3070",
23
+ "init_seed": 3,
24
+ "split_seed": 42
25
+ }
virchow2_abmil_focal_maskgate_5seed/seed3/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_abmil_focal_maskgate_5seed/seed3",
3
+ "best_epoch": 44,
4
+ "best_val_qwk": 0.9141272890088641,
5
+ "n_parameters": 1479815,
6
+ "weights_sha256": "fd95421639858940513c054f5116a233c29ee449357c5db9fe119ca14bb2ef7c",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_abmil_focal_maskgate_5seed/seed4/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eda8dcc1f5b36bcab163930ba2b4d69e0b1a1215aad78f234ea8b11076da7c07
3
+ size 5920172
virchow2_abmil_focal_maskgate_5seed/seed4/config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_abmil_gated_seed4",
3
+ "backbone": "virchow2 (1280d)",
4
+ "gating": true,
5
+ "gating_mode": "v2",
6
+ "dropout": 0.25,
7
+ "patch_dropout": 0.2,
8
+ "early_stopping_patience": 0,
9
+ "cosine_lr": false,
10
+ "epochs": 50,
11
+ "learning_rate": 0.0001,
12
+ "weight_decay": 0.01,
13
+ "loss_type": "focal",
14
+ "focal_gamma": 2.0,
15
+ "accumulation_steps": 8,
16
+ "transmil_layers": "N/A",
17
+ "variance_threshold": 122.0,
18
+ "z_score_factor": 2.0,
19
+ "train_slides": 7961,
20
+ "val_slides": 1592,
21
+ "test_slides": 1061,
22
+ "parameters": 1479815,
23
+ "device": "NVIDIA GeForce RTX 3070",
24
+ "init_seed": 4,
25
+ "split_seed": 42
26
+ }
virchow2_abmil_focal_maskgate_5seed/seed4/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_abmil_focal_maskgate_5seed/seed4",
3
+ "best_epoch": 33,
4
+ "best_val_qwk": 0.9137856059183678,
5
+ "n_parameters": 1479815,
6
+ "weights_sha256": "eda8dcc1f5b36bcab163930ba2b4d69e0b1a1215aad78f234ea8b11076da7c07",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_fullgate_5fold/fold_1/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09e1ae98c9a5bb1d322fca1b90a6e9cc0d1b22898557a955e2356402ac831517
3
+ size 163340024
virchow2_transmil_coral_fullgate_5fold/fold_1/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_v2_fullgate_5fold_coral",
3
+ "fold": 1,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 0,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 0,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_fullgate_5fold/fold_1/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_fullgate_5fold/fold_1",
3
+ "best_epoch": 41,
4
+ "best_val_qwk": 0.9253148951135173,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "09e1ae98c9a5bb1d322fca1b90a6e9cc0d1b22898557a955e2356402ac831517",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_fullgate_5fold/fold_2/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3bbaddb7eba57549fff222daf0d32e0a0b4955f0fc841904e9866a0eb54b9232
3
+ size 163340024
virchow2_transmil_coral_fullgate_5fold/fold_2/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_v2_fullgate_5fold_coral",
3
+ "fold": 2,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 0,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 1,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_fullgate_5fold/fold_2/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_fullgate_5fold/fold_2",
3
+ "best_epoch": 47,
4
+ "best_val_qwk": 0.9172772870217861,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "3bbaddb7eba57549fff222daf0d32e0a0b4955f0fc841904e9866a0eb54b9232",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_fullgate_5fold/fold_3/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc339930a718193945d10c139a58e4e9708cddbcbb971e1482088d683326e798
3
+ size 163340024
virchow2_transmil_coral_fullgate_5fold/fold_3/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_v2_fullgate_5fold_coral",
3
+ "fold": 3,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 0,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 2,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_fullgate_5fold/fold_3/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_fullgate_5fold/fold_3",
3
+ "best_epoch": 46,
4
+ "best_val_qwk": 0.9271063800130852,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "fc339930a718193945d10c139a58e4e9708cddbcbb971e1482088d683326e798",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_fullgate_5fold/fold_4/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22cf54253fdc87852bc4baed158104df1db3985593ff1cc80731960fcb448468
3
+ size 163340024
virchow2_transmil_coral_fullgate_5fold/fold_4/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_v2_fullgate_5fold_coral",
3
+ "fold": 4,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 0,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 3,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_fullgate_5fold/fold_4/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_fullgate_5fold/fold_4",
3
+ "best_epoch": 46,
4
+ "best_val_qwk": 0.9169733837999474,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "22cf54253fdc87852bc4baed158104df1db3985593ff1cc80731960fcb448468",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_fullgate_5fold/fold_5/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b543aad365cc89d7cee8b42322ccf357572fc1248d87cf13161298c72a321b04
3
+ size 163340024
virchow2_transmil_coral_fullgate_5fold/fold_5/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_v2_fullgate_5fold_coral",
3
+ "fold": 5,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 0,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8492,
23
+ "val_slides": 2122,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 4,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_fullgate_5fold/fold_5/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_fullgate_5fold/fold_5",
3
+ "best_epoch": 47,
4
+ "best_val_qwk": 0.9269270799228082,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "b543aad365cc89d7cee8b42322ccf357572fc1248d87cf13161298c72a321b04",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_maskgate_5fold/fold_1/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5fb6f6ae4232ca123faaf8f3fd21222c72019aa505500824611806f939d34a17
3
+ size 163340024
virchow2_transmil_coral_maskgate_5fold/fold_1/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_gated_5fold_coral",
3
+ "fold": 1,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 10,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 0,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_maskgate_5fold/fold_1/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_maskgate_5fold/fold_1",
3
+ "best_epoch": 22,
4
+ "best_val_qwk": 0.9217563225587565,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "5fb6f6ae4232ca123faaf8f3fd21222c72019aa505500824611806f939d34a17",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_maskgate_5fold/fold_2/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b88e12854d5ef769385b8ea5c51c50ce737a316a3a24172c28361d30d73e3d7
3
+ size 163340024
virchow2_transmil_coral_maskgate_5fold/fold_2/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_gated_5fold_coral",
3
+ "fold": 2,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 10,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 1,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_maskgate_5fold/fold_2/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_maskgate_5fold/fold_2",
3
+ "best_epoch": 28,
4
+ "best_val_qwk": 0.9186392839205172,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "5b88e12854d5ef769385b8ea5c51c50ce737a316a3a24172c28361d30d73e3d7",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_maskgate_5fold/fold_3/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe2ab2e7064754a491720d9d243d762b3066e252bd22c69e2a82e86e336e50ea
3
+ size 163340024
virchow2_transmil_coral_maskgate_5fold/fold_3/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_gated_5fold_coral",
3
+ "fold": 3,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 10,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 2,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_maskgate_5fold/fold_3/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_maskgate_5fold/fold_3",
3
+ "best_epoch": 36,
4
+ "best_val_qwk": 0.927293241921137,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "fe2ab2e7064754a491720d9d243d762b3066e252bd22c69e2a82e86e336e50ea",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_maskgate_5fold/fold_4/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a81d1711f03b1fb0b9c8d1b05b99f7f6ff314f9f502352c7057d4e7557f50a5
3
+ size 163340024
virchow2_transmil_coral_maskgate_5fold/fold_4/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_gated_5fold_coral",
3
+ "fold": 4,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 10,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8491,
23
+ "val_slides": 2123,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 3,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_maskgate_5fold/fold_4/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_maskgate_5fold/fold_4",
3
+ "best_epoch": 28,
4
+ "best_val_qwk": 0.9183432329171636,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "9a81d1711f03b1fb0b9c8d1b05b99f7f6ff314f9f502352c7057d4e7557f50a5",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_coral_maskgate_5fold/fold_5/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2b0756806725789966e22255388578d3f7fb32f78c5fb680751db409efd3d14
3
+ size 163340024
virchow2_transmil_coral_maskgate_5fold/fold_5/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "virchow2_transmil_gated_5fold_coral",
3
+ "fold": 5,
4
+ "n_folds": 5,
5
+ "backbone": "virchow2 (1280d)",
6
+ "aggregator": "transmil",
7
+ "gating": true,
8
+ "gating_mode": "v2",
9
+ "dropout": 0.4,
10
+ "patch_dropout": 0.2,
11
+ "early_stopping_patience": 10,
12
+ "cosine_lr": true,
13
+ "epochs": 50,
14
+ "learning_rate": 1e-05,
15
+ "weight_decay": 0.01,
16
+ "loss_type": "coral",
17
+ "focal_gamma": "N/A",
18
+ "accumulation_steps": 8,
19
+ "transmil_layers": 2,
20
+ "variance_threshold": 122.0,
21
+ "z_score_factor": 2.0,
22
+ "train_slides": 8492,
23
+ "val_slides": 2122,
24
+ "device": "NVIDIA GeForce RTX 3070",
25
+ "fold_seed": 4,
26
+ "split_seed": 42
27
+ }
virchow2_transmil_coral_maskgate_5fold/fold_5/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_coral_maskgate_5fold/fold_5",
3
+ "best_epoch": 36,
4
+ "best_val_qwk": 0.9268811187789894,
5
+ "n_parameters": 40834054,
6
+ "weights_sha256": "f2b0756806725789966e22255388578d3f7fb32f78c5fb680751db409efd3d14",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_focal_fullgate_5seed/seed0/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df777ea7f2ea0ae71c7ffad59de84a8d478536011e648e6c94e15294508f3bc0
3
+ size 163342588
virchow2_transmil_focal_fullgate_5seed/seed0/config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "experiment_name": "gated_transmil_v2_fullgate_seed0",
3
+ "backbone": "virchow2 (1280d)",
4
+ "gating": true,
5
+ "gating_mode": "v2",
6
+ "dropout": 0.4,
7
+ "patch_dropout": 0.2,
8
+ "early_stopping_patience": 0,
9
+ "cosine_lr": true,
10
+ "epochs": 50,
11
+ "learning_rate": 1e-05,
12
+ "weight_decay": 0.01,
13
+ "loss_type": "focal",
14
+ "focal_gamma": 2.0,
15
+ "accumulation_steps": 8,
16
+ "transmil_layers": 2,
17
+ "variance_threshold": 122.0,
18
+ "z_score_factor": 2.0,
19
+ "train_slides": 7961,
20
+ "val_slides": 1592,
21
+ "test_slides": 1061,
22
+ "parameters": 40834695,
23
+ "device": "NVIDIA GeForce RTX 3070",
24
+ "init_seed": 0,
25
+ "split_seed": 42
26
+ }
virchow2_transmil_focal_fullgate_5seed/seed0/release_info.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "virchow2_transmil_focal_fullgate_5seed/seed0",
3
+ "best_epoch": 50,
4
+ "best_val_qwk": 0.9240166377712489,
5
+ "n_parameters": 40834695,
6
+ "weights_sha256": "df777ea7f2ea0ae71c7ffad59de84a8d478536011e648e6c94e15294508f3bc0",
7
+ "conversion_check": {
8
+ "random_bags": 3,
9
+ "max_abs_logit_diff": 0.0
10
+ }
11
+ }
virchow2_transmil_focal_fullgate_5seed/seed1/best_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2702fb1364b2da41fdb35eaf93e322762fb0c3b2c4f7a8a8fe229c9250fee72
3
+ size 163342588