renable model training widget

.gitignore

@@ -48,3 +48,5 @@ fibsem/config/*.yaml
 !fibsem/config/microscope-configuration.yaml
 !fibsem/config/user-configurations.yaml
 !fibsem/config/tescan_manipulator.yaml
+
+**/wandb/*

fibsem/segmentation/config.yml

@@ -1,12 +1,26 @@
-data_path:  /home/patrick/github/data/liftout/training/train/images 
-label_path: /home/patrick/github/data/liftout/training/train/labels 
-save_path: /home/patrick/github/fibsem/fibsem/segmentation/models/autoliftout/xl/base/nc3/earlystop
-checkpoint: null 
-encoder: "resnet50" 
-epochs: 5
-batch_size: 4 
-num_classes: 3 
-lr: 3.0e-4
-wandb: true 
-wandb_project: "autoliftout"
-wandb_entity: "demarcolab"
+# data
+data_paths:  [/path/to/data, /path/to/second/data]                  # paths to image data (multiple supported)
+label_paths: [/path/to/data/labels, /path/to/second/data/labels]    # paths to label data (multiple supported)
+save_path: /path/to/save/checkpoints                                # path to save checkpoints (checkpointed each epoch)
+checkpoint: null                                                    # checkpoint to resume from
+
+# model
+encoder: "resnet34"                             # segmentation model encoder (imagenet)
+num_classes: 6                                  # number of classes
+
+# training
+epochs: 50                                      # number of epochs
+split: 0.1                                      # train / val split
+batch_size: 4                                   # batch size
+lr: 3.0e-4                                      # initial learning rate
+
+# logging
+train_log_freq: 32                              # frequency to log training images
+val_log_freq: 32                                # frequency to log validation images
+
+# wandb
+wandb: true                                     # enable wandb logging
+wandb_project: "autolamella-mega"               # wandb project
+wandb_entity: "openfibsem"                      # wandb user / org
+model_type: "mega-model"                        # model type note (descriptive only)
+note: "notes about this specific training run"  # additional trianing note (descriptive only)

fibsem/ui/FibsemLabellingUI.py

@@ -24,7 +24,7 @@
 # setup a basic logger
 logging.basicConfig(stream=sys.stdout, level=logging.INFO)
 
-_DEBUG = True
+_DEBUG = False
 
 # new line char in html
 NL = "<br>"

fibsem/ui/FibsemModelTrainingWidget.py

@@ -51,9 +51,9 @@ def setup_connections(self):
 
         checkpoint = config["checkpoint"] if config["checkpoint"] is not None else ""
 
-        self.lineEdit_data_path.setText(config["data_path"])
-        self.lineEdit_label_path.setText(config["label_path"])
-        self.lineEdit_save_path.setText(config["path"])
+        self.lineEdit_data_path.setText(config["data_paths"][0])
+        self.lineEdit_label_path.setText(config["label_paths"][0])
+        self.lineEdit_save_path.setText(config["save_path"])
         self.comboBox_encoder.setCurrentText(config["encoder"])
         self.lineEdit_checkpoint.setText(checkpoint)
         self.spinBox_num_classes.setValue(config["num_classes"])
@@ -69,9 +69,9 @@ def train_model(self):
         checkpoint = None if checkpoint == "" else checkpoint
 
         config = {
-            "data_path": self.lineEdit_data_path.text(),
-            "label_path": self.lineEdit_label_path.text(),
-            "path": self.lineEdit_save_path.text(),
+            "data_paths": [self.lineEdit_data_path.text()],
+            "label_paths": [self.lineEdit_label_path.text()],
+            "save_path": self.lineEdit_save_path.text(),
             "encoder": self.comboBox_encoder.currentText(),
             "checkpoint": checkpoint,
             "num_classes": int(self.spinBox_num_classes.value()),
@@ -113,10 +113,7 @@ def train_worker(self, config: dict):
             optimizer,
             train_data_loader,
             val_data_loader,
-            epochs=config["epochs"],
-            save_dir=config["save_path"],
-            WANDB=config["wandb"],
-            ui=self.ui_signal,
+            config=config,
         )
 
     def train_model_yielded(self, info: dict):