new files added

MPI-Dortmund · Jul 31, 2024 · 3568e80 · 3568e80
1 parent 6d2973a
commit 3568e80
Show file tree

Hide file tree

Showing 13 changed files with 2,027 additions and 13 deletions.
diff --git a/VAE.ipynb b/VAE.ipynb
diff --git a/noskip32.ipynb b/noskip32.ipynb
diff --git a/optuna.ipynb b/optuna.ipynb
diff --git a/optuna_study.db b/optuna_study.db
@@ -0,0 +1,40 @@
+{"op_code": 0, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "study_name": "3D_UNet_Optimization", "directions": [1]}
+{"op_code": 4, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "study_id": 0, "datetime_start": "2024-07-25T11:42:07.335452"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 0, "param_name": "lr", "param_value_internal": 6.896421114751126e-05, "distribution": "{\"name\": \"FloatDistribution\", \"attributes\": {\"step\": null, \"low\": 1e-05, \"high\": 0.1, \"log\": true}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 0, "param_name": "optimizer", "param_value_internal": 0, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [\"Adam\", \"SGD\", \"RMSprop\"]}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 0, "param_name": "filters", "param_value_internal": 0, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [32, 64]}}"}
+{"op_code": 6, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 0, "state": 3, "values": null, "datetime_complete": "2024-07-25T11:42:07.372180"}
+{"op_code": 4, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "study_id": 0, "datetime_start": "2024-07-25T11:42:38.184586"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 1, "param_name": "lr", "param_value_internal": 0.004413506973711808, "distribution": "{\"name\": \"FloatDistribution\", \"attributes\": {\"step\": null, \"low\": 1e-05, \"high\": 0.1, \"log\": true}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 1, "param_name": "optimizer", "param_value_internal": 2, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [\"Adam\", \"SGD\", \"RMSprop\"]}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 1, "param_name": "filters", "param_value_internal": 1, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [32, 64]}}"}
+{"op_code": 6, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 1, "state": 3, "values": null, "datetime_complete": "2024-07-25T11:42:40.221492"}
+{"op_code": 4, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "study_id": 0, "datetime_start": "2024-07-25T11:43:13.900783"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "param_name": "lr", "param_value_internal": 0.009858907638988407, "distribution": "{\"name\": \"FloatDistribution\", \"attributes\": {\"step\": null, \"low\": 1e-05, \"high\": 0.1, \"log\": true}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "param_name": "optimizer", "param_value_internal": 2, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [\"Adam\", \"SGD\", \"RMSprop\"]}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "param_name": "filters", "param_value_internal": 0, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [32, 64]}}"}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "step": 0, "intermediate_value": 0.012406982839193618}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "step": 1, "intermediate_value": 0.012296284513483437}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "step": 2, "intermediate_value": 0.012192022176946935}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "step": 3, "intermediate_value": 0.01325888847546015}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "step": 4, "intermediate_value": 0.01175757600621331}
+{"op_code": 6, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 2, "state": 1, "values": [0.01175757600621331], "datetime_complete": "2024-07-25T12:59:25.440510"}
+{"op_code": 4, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "study_id": 0, "datetime_start": "2024-07-25T12:59:25.447681"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "param_name": "lr", "param_value_internal": 0.00375852733206065, "distribution": "{\"name\": \"FloatDistribution\", \"attributes\": {\"step\": null, \"low\": 1e-05, \"high\": 0.1, \"log\": true}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "param_name": "optimizer", "param_value_internal": 0, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [\"Adam\", \"SGD\", \"RMSprop\"]}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "param_name": "filters", "param_value_internal": 0, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [32, 64]}}"}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "step": 0, "intermediate_value": 0.012742554306107707}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "step": 1, "intermediate_value": 0.01393519094331604}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "step": 2, "intermediate_value": 0.013373396859550501}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "step": 3, "intermediate_value": 0.012453339947102671}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "step": 4, "intermediate_value": 0.012303508424347635}
+{"op_code": 6, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 3, "state": 1, "values": [0.012303508424347635], "datetime_complete": "2024-07-25T14:11:46.123018"}
+{"op_code": 4, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "study_id": 0, "datetime_start": "2024-07-25T14:11:46.130124"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "param_name": "lr", "param_value_internal": 0.08937778989479532, "distribution": "{\"name\": \"FloatDistribution\", \"attributes\": {\"step\": null, \"low\": 1e-05, \"high\": 0.1, \"log\": true}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "param_name": "optimizer", "param_value_internal": 1, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [\"Adam\", \"SGD\", \"RMSprop\"]}}"}
+{"op_code": 5, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "param_name": "filters", "param_value_internal": 1, "distribution": "{\"name\": \"CategoricalDistribution\", \"attributes\": {\"choices\": [32, 64]}}"}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "step": 0, "intermediate_value": 0.01562090049086887}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "step": 1, "intermediate_value": 0.248305968297006}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "step": 2, "intermediate_value": 0.21679665308311472}
+{"op_code": 7, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "step": 3, "intermediate_value": 0.05948987290931753}
+{"op_code": 6, "worker_id": "0bc2c3c2-d18c-471c-8b1c-f925a91155e2-135722348189504", "trial_id": 4, "state": 3, "values": null, "datetime_complete": "2024-07-25T16:16:18.517776"}
diff --git a/resources/configs/config_unet.json b/resources/configs/config_unet.json
@@ -0,0 +1,19 @@
+{
+  "identifier": "UNet",
+  "network_config": {},
+
+  "train_config":{
+    "loss": "TripletLoss",
+    "tl_margin": 0.539,
+    "miner": true,
+    "miner_margin": 0.734,
+    "learning_rate": 5.945e-05,
+    "optimizer": "Adam",
+    "weight_decay": 0,
+    "batchsize": 64,
+    "patience": 50,
+    "aug_train_shift_distance": 2
+  },
+
+  "distance": "COSINE"
+}
diff --git a/resources/configs/config_unet_GN.json b/resources/configs/config_unet_GN.json
@@ -0,0 +1,19 @@
+{
+  "identifier": "UNet_GN",
+  "network_config": {},
+
+  "train_config":{
+    "loss": "TripletLoss",
+    "tl_margin": 0.539,
+    "miner": true,
+    "miner_margin": 0.734,
+    "learning_rate": 5.945e-05,
+    "optimizer": "Adam",
+    "weight_decay": 0,
+    "batchsize": 64,
+    "patience": 50,
+    "aug_train_shift_distance": 2
+  },
+
+  "distance": "COSINE"
+}
diff --git a/tomotwin/modules/common/io/mrc_format.py b/tomotwin/modules/common/io/mrc_format.py
@@ -19,8 +19,8 @@ class MrcFormat:
     def read(pth: str) -> np.array:
         vol = None
         try:
-            with mrcfile.open(pth, mode="r", permissive=True) as mrc:
-                vol = mrc.data
+            with mrcfile.open(pth, permissive=True) as mrc:
+                vol = mrc.data.astype(np.float32)
         except ValueError as e:
             raise Exception(f"Failed reading {pth}") from e
         return vol
diff --git a/tomotwin/modules/common/preprocess.py b/tomotwin/modules/common/preprocess.py
@@ -29,3 +29,9 @@ def label_filename(path: str) -> str:
 def norm(vol: np.array) -> np.array:
     """Applies standard normalization"""
     return (vol-np.mean(vol))/np.std(vol)
+
+def norm2(vol: np.array) -> np.array:
+    min_val = np.min(vol)
+    max_val = np.max(vol)
+    normalized_vol = (vol - min_val) / (max_val - min_val)
+    return normalized_vol
diff --git a/tomotwin/modules/training/mrctriplethandler.py b/tomotwin/modules/training/mrctriplethandler.py
@@ -27,7 +27,7 @@ class MRCTripletHandler(TripletHandler):
     @staticmethod
     def read_mrc_and_norm(pth: str) -> np.array:
         vol = MrcFormat.read(pth)
-        vol = preprocess.norm(vol)
+        vol = preprocess.norm2(vol)
         return vol
 
     def handle(self, triplet: FilePathTriplet) -> NumpyTriplet:

diff --git a/tomotwin/modules/training/torchtrainer.py b/tomotwin/modules/training/torchtrainer.py
@@ -258,14 +258,16 @@ def run_batch(self, batch: Dict):
         :param batch: Dictionary with batch data
         :return: Loss of the batch
         """
-        anchor_vol = batch["anchor"].to(self.device, non_blocking=True)
-        positive_vol = batch["positive"].to(self.device, non_blocking=True)
-        negative_vol = batch["negative"].to(self.device, non_blocking=True)
-        full_input = torch.cat((anchor_vol,positive_vol,negative_vol), dim=0)
+        #anchor_vol = batch["anchor"].to(self.device, non_blocking=True)
+        #positive_vol = batch["positive"].to(self.device, non_blocking=True)
+        #negative_vol = batch["negative"].to(self.device, non_blocking=True)
+        anchor_vol = batch["anchor"]
+        positive_vol = batch["positive"]
+        negative_vol = batch["negative"]
+        full_input = torch.cat((anchor_vol,positive_vol,negative_vol), dim=0).to(self.device, non_blocking=True)
         with autocast():
             out = self.model.forward(full_input)
             out = torch.split(out, anchor_vol.shape[0], dim=0)
-
             loss = self.criterion(
                 out[0],
                 out[1],