Implement evaluation against mimic reports in test set evaluation

Author: ttanida

src/full_model/custom_dataset.py

@@ -32,7 +32,7 @@ def __getitem__(self, index):
                 # we only need the reference phrases when computing e.g. BLEU during evaluation
                 bbox_phrases = self.tokenized_dataset[index]["bbox_phrases"]  # List[str]
 
-                # we need the study_id to retrieve the corresponding mimic_cxr report during evaliation
+                # we need the study_id to retrieve the corresponding mimic_cxr report during evaluation
                 study_id = self.tokenized_dataset[index]["study_id"]  # str
 
             # cv2.imread by default loads an image with 3 channels

src/full_model/evaluate_full_model/evaluate_model.py

@@ -224,7 +224,7 @@ def compute_intersection_and_union_area_per_region(detections, targets, class_de
     obj_detector_scores["sum_union_area_per_region"] += union_area_per_region_batch
 
 
-def get_val_losses_and_other_metrics(model, val_dl, log_file, epoch):
+def get_val_losses_and_other_metric_scores(model, val_dl, log_file, epoch):
     """
     Args:
         model (nn.Module): The input model to be evaluated.
@@ -474,7 +474,7 @@ def evaluate_model(model, train_losses_dict, val_dl, lr_scheduler, optimizer, sc
         obj_detector_scores,
         region_selection_scores,
         region_abnormal_scores,
-    ) = get_val_losses_and_other_metrics(model, val_dl, log_file, epoch)
+    ) = get_val_losses_and_other_metric_scores(model, val_dl, log_file, epoch)
 
     # TODO: delete 2nd and 3rd condition (since they are only there to save time)
     if not PRETRAIN_WITHOUT_LM_MODEL and overall_steps_taken > 25000 and bool_evaluate_language_model: