Merge branch 'master' into hotfix/DetectionRandomAffine-fix-target-size

src/super_gradients/common/auto_logging/auto_logger.py

@@ -52,7 +52,7 @@ def _setup_logging(self, filename: str, copy_already_logged_messages: bool, file
 
         if copy_already_logged_messages and self.filename is not None and os.path.exists(self.filename):
             with open(self.filename, "r", encoding="utf-8") as src:
-                with open(filename, "w") as dst:
+                with open(filename, "w", encoding="utf-8") as dst:
                     dst.write(src.read())
 
         file_logging_level = log_level or env_variables.FILE_LOG_LEVEL

src/super_gradients/common/plugins/deci_client.py

@@ -161,7 +161,7 @@ def load_code_from_zipfile(*, file: str, target_path: str, package_name: str = "
             zipfile.extractall(package_path)
 
         # add an init file that imports all code files
-        with open(os.path.join(package_path, "__init__.py"), "w") as init_file:
+        with open(os.path.join(package_path, "__init__.py"), "w", encoding="utf-8") as init_file:
             all_str = "\n\n__all__ = ["
             for code_file in os.listdir(path=package_path):
                 if code_file.endswith(".py") and not code_file.startswith("__init__"):

src/super_gradients/common/sg_loggers/base_sg_logger.py

@@ -156,7 +156,7 @@ def _init_log_file(self):
 
     @multi_process_safe
     def _write_to_log_file(self, lines: list):
-        with open(self.experiment_log_path, "a" if os.path.exists(self.experiment_log_path) else "w") as log_file:
+        with open(self.experiment_log_path, "a" if os.path.exists(self.experiment_log_path) else "w", encoding="utf-8") as log_file:
             for line in lines:
                 log_file.write(line + "\n")
 
@@ -345,7 +345,7 @@ def _save_code(self):
                 name = name + ".py"
 
             path = os.path.join(self._local_dir, name)
-            with open(path, "w") as f:
+            with open(path, "w", encoding="utf-8") as f:
                 f.write(code)
 
             self.add_file(name)

src/super_gradients/convert_recipe_to_code.py

@@ -311,7 +311,7 @@ def main():
         key_to_replace_with = f"{key}"
         content = content.replace(key_to_search, key_to_replace_with)
 
-    with open(output_script_path, "w") as f:
+    with open(output_script_path, "w", encoding="utf-8") as f:
         black = try_import_black()
         if black is not None:
             content = black.format_str(content, mode=black.FileMode(line_length=160))

src/super_gradients/training/datasets/detection_datasets/pascal_voc_detection.py

@@ -166,7 +166,7 @@ def get_coord(box_coord):
                     xmin, ymin, xmax, ymax = get_coord("xmin"), get_coord("ymin"), get_coord("xmax"), get_coord("ymax")
                     labels.append(" ".join([xmin, ymin, xmax, ymax, str(PASCAL_VOC_2012_CLASSES_LIST.index(cls))]))
 
-            with open(new_label_path, "w") as f:
+            with open(new_label_path, "w", encoding="utf-8") as f:
                 f.write("\n".join(labels))
 
         urls = [

src/super_gradients/training/datasets/samplers/class_balanced_sampler.py

@@ -100,7 +100,7 @@ def precompute_sample_repeat_factors(
 
         str_repeat_factors = [np.format_float_positional(rf, trim="0", precision=4) for rf in repeat_factors]
 
-        with open(output_path, "w") as f:
+        with open(output_path, "w", encoding="utf-8") as f:
             json.dump(str_repeat_factors, f)
 
     @staticmethod

src/super_gradients/training/sg_trainer/sg_trainer.py

@@ -679,6 +679,15 @@ def _save_checkpoint(
             train_metrics_titles = get_metrics_titles(self.train_metrics)
             all_metrics["train"] = {metric_name: float(train_metrics_dict[metric_name]) for metric_name in train_metrics_titles}
 
+        best_checkpoint = (curr_tracked_metric > self.best_metric and self.greater_metric_to_watch_is_better) or (
+            curr_tracked_metric < self.best_metric and not self.greater_metric_to_watch_is_better
+        )
+
+        if best_checkpoint:
+            # STORE THE CURRENT metric AS BEST
+            self.best_metric = curr_tracked_metric
+            self._best_ckpt_metrics = all_metrics
+
         # BUILD THE state_dict
         state = {
             "net": unwrap_model(self.net).state_dict(),
@@ -713,13 +722,7 @@ def _save_checkpoint(
             self.sg_logger.add_checkpoint(tag=f"ckpt_epoch_{epoch}.pth", state_dict=state, global_step=epoch)
 
         # OVERRIDE THE BEST CHECKPOINT AND best_metric IF metric GOT BETTER THAN THE PREVIOUS BEST
-        if (curr_tracked_metric > self.best_metric and self.greater_metric_to_watch_is_better) or (
-            curr_tracked_metric < self.best_metric and not self.greater_metric_to_watch_is_better
-        ):
-            # STORE THE CURRENT metric AS BEST
-            self.best_metric = curr_tracked_metric
-
-            self._best_ckpt_metrics = all_metrics
+        if best_checkpoint:
             self.sg_logger.add_checkpoint(tag=self.ckpt_best_name, state_dict=state, global_step=epoch)
 
             # RUN PHASE CALLBACKS

tests/unit_tests/export_detection_model_test.py

@@ -568,7 +568,7 @@ def manual_test_export_export_all_variants(self):
         os.makedirs(export_dir, exist_ok=True)
 
         benchmark_command_dir = "benchmark_command.sh"
-        with open(benchmark_command_dir, "w") as f:
+        with open(benchmark_command_dir, "w", encoding="utf-8") as f:
             pass
 
         for output_predictions_format in [DetectionOutputFormatMode.BATCH_FORMAT, DetectionOutputFormatMode.FLAT_FORMAT]:

tests/unit_tests/pose_estimation_metrics_test.py

@@ -87,7 +87,7 @@ def convert_predictions_to_target_format(preds) -> List[PoseEstimationPrediction
         with tempfile.TemporaryDirectory() as td:
             res_file = os.path.join(td, "keypoints_coco2017_results.json")
 
-            with open(res_file, "w") as f:
+            with open(res_file, "w", encoding="utf-8") as f:
                 json.dump(coco_pred, f, sort_keys=True, indent=4)
 
             coco_dt = self._load_coco_groundtruth(with_crowd, with_duplicates, with_invisible_keypoitns)

utils_script/create_sub_coco.py

@@ -46,7 +46,7 @@ def _copy_to_new_dir(mode: str, n_images: int, input_data_dir: Path, dest_data_d
     dest_images_dir = dest_data_dir / "images" / f"{mode}2017"
     dest_images_dir.mkdir(exist_ok=True, parents=True)
 
-    with open(dest_instances_path, "w") as f:
+    with open(dest_instances_path, "w", encoding="utf-8") as f:
         json.dump(instances, f)
 
     for image_name in kept_images_name: