realization of metrics

user-friendly-metrics.py

@@ -190,10 +190,19 @@ def calculate_from_payload(payload: dict,
         print("models: ", models)
         print("sequence_list: ", sequence_list)
 
-    output = {}
+    metrics_per_sequence = {}
+    metrics_global = {}
+    for model in models:
+        metrics_global[model] = {}
+        metrics_global[model]["all"] = {}
+        for filter, filter_ranges in filters.items():
+            metrics_global[model][filter] = {}
+            for filter_range in filter_ranges:
+                filter_range_name = filter_range[0]
+                metrics_global[model][filter][filter_range_name] = {}
 
     for sequence in sequence_list:
-        output[sequence] = {}
+        metrics_per_sequence[sequence] = {}
         frames = payload['sequences'][sequence][gt_field_name]
         
         all_formated_references = {"all": []}
@@ -216,7 +225,6 @@ def calculate_from_payload(payload: dict,
                         if filter_value >= filter_range_limits[0] and filter_value <= filter_range_limits[1]:
                             all_formated_references[filter][filter_range_name].append([frame_id+1, index, x, y, w, h])
 
-
         for model in models:
             frames = payload['sequences'][sequence][model]
             formated_predictions = []
@@ -233,24 +241,33 @@ def calculate_from_payload(payload: dict,
                 print("formated_predictions: ", formated_predictions)
                 print("formated_references: ", all_formated_references)
             if len(formated_predictions) == 0:
-                output[sequence][model] = "Model had no predictions."
+                metrics_per_sequence[sequence][model] = "Model had no predictions."
             elif len(all_formated_references["all"]) == 0:
-                output[sequence][model] = "No ground truth."
+                metrics_per_sequence[sequence][model] = "No ground truth."
             else:
-                output[sequence][model] = {}
-                output[sequence][model]["all"] = calculate(formated_predictions, all_formated_references["all"], max_iou=max_iou, recognition_thresholds = recognition_thresholds)
+                metrics_per_sequence[sequence][model] = {}
+
+                sequence_metrics = calculate(formated_predictions, all_formated_references["all"], max_iou=max_iou, recognition_thresholds = recognition_thresholds)
+                sequence_metrics = realize_metrics(sequence_metrics, recognition_thresholds)
+                metrics_per_sequence[sequence][model]["all"] = sequence_metrics
+
+                metrics_global[model]["all"] = sum_dicts(metrics_global[model], sequence_metrics)
+                metrics_global[model]["all"] = realize_metrics(metrics_global[model]["all"], recognition_thresholds)
+                                
                 for filter, filter_ranges in filters.items():
-                    output[sequence][model][filter] = {}
+                    metrics_per_sequence[sequence][model][filter] = {}
                     for filter_range in filter_ranges:
+                        
                         filter_range_name = filter_range[0]
-                        output[sequence][model][filter][filter_range_name] = calculate(formated_predictions, all_formated_references[filter][filter_range_name], max_iou=max_iou, recognition_thresholds = recognition_thresholds)
+                        sequence_metrics = calculate(formated_predictions, all_formated_references[filter][filter_range_name], max_iou=max_iou, recognition_thresholds = recognition_thresholds)
+                        sequence_metrics = realize_metrics(sequence_metrics, recognition_thresholds)
+                        metrics_per_sequence[sequence][model][filter][filter_range_name] = sequence_metrics
 
-    global_and_per_sequence_output = {}
-    global_and_per_sequence_output["global"] = per_sequence_to_global(output)
-    global_and_per_sequence_output["per_sequence"] = output
+                        metrics_global[model][filter][filter_range_name] = sum_dicts(metrics_global[model][filter][filter_range_name], sequence_metrics)
+                        metrics_global[model][filter][filter_range_name] = realize_metrics(metrics_global[model][filter][filter_range_name], recognition_thresholds)
     
-    return global_and_per_sequence_output
-            
+    return metrics_per_sequence
+
 def sum_dicts(dict1, dict2):
     """
     Recursively sums the numerical values in two nested dictionaries.
@@ -272,13 +289,13 @@ def sum_dicts(dict1, dict2):
 
 def realize_metrics(metrics_dict, 
                    recognition_thresholds):
-    print(metrics_dict)
+    
     metrics_dict["precision"] = metrics_dict["tp"]/(metrics_dict["tp"]+metrics_dict["fp"])
     metrics_dict["recall"] = metrics_dict["tp"]/(metrics_dict["tp"]+metrics_dict["fn"])
     metrics_dict["f1"] = 2*metrics_dict["precision"]*metrics_dict["recall"]/(metrics_dict["precision"]+metrics_dict["recall"])
 
     for th in recognition_thresholds:
-        metrics_dict[f"recognition_{th}"] = metrics_dict[f"recognized_{th}"]/metrics_dict["num_gt_ids"]
+        metrics_dict[f"recognition_{th}"] = metrics_dict[f"recognized_{th}"]/metrics_dict["unique_gt_ids"]
         
     return metrics_dict