Merge back 1.2.0rc1

.github/workflows/code_scan.yml

@@ -49,7 +49,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: 3.10
+          python-version: "3.10"
       - name: Install dependencies
         run: python -m pip install tox
       - name: Bandit Scanning
@@ -71,7 +71,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: 3.10
+          python-version: "3.10"
       - name: Install dependencies
         run: python -m pip install -r requirements/dev.txt
       - name: Snyk Scanning

otx/algorithms/action/configs/classification/base/__init__.py

@@ -0,0 +1,4 @@
+"""Configs Initialization of OTX Action Classification Tasks."""
+
+# Copyright (C) 2021-2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0

otx/algorithms/common/tasks/nncf_task.py

@@ -162,7 +162,7 @@ def _prepare_optimize(self):
         # last batch size of 1 causes undefined behaviour for batch normalization
         # when initializing and training NNCF
         if self._data_cfg is not None:
-            data_loader = self._recipe_cfg.data.get("train_dataloader", {})
+            data_loader = self._recipe_cfg.data.get("train_dataloader", ConfigDict())
             samples_per_gpu = data_loader.get("samples_per_gpu", self._recipe_cfg.data.get("samples_per_gpu"))
             otx_dataset = get_configs_by_keys(self._data_cfg.data.train, "otx_dataset")
             assert len(otx_dataset) == 1

otx/algorithms/detection/adapters/mmdet/datasets/tiling.py

@@ -96,25 +96,6 @@ def __init__(
 
         self.dataset = dataset
         self.tiles = self.gen_tile_ann()
-        self.cache_tiles()
-
-    @timeit
-    def cache_tiles(self):
-        """Cache tiles to disk."""
-        pbar = tqdm(total=len(self.tiles))
-        pre_img_idx = None
-        for i, tile in enumerate(self.tiles):
-            tile["tile_path"] = osp.join(
-                self.tmp_folder, "_".join([str(i), tile["uuid"], tile["ori_filename"], ".jpg"])
-            )
-            x_1, y_1, x_2, y_2 = tile["tile_box"]
-            dataset_idx = tile["dataset_idx"]
-            if dataset_idx != pre_img_idx:
-                ori_img = self.dataset[dataset_idx]["img"]
-                pre_img_idx = dataset_idx
-
-            mmcv.imwrite(ori_img[y_1:y_2, x_1:x_2, :], tile["tile_path"])
-            pbar.update(1)
 
     @timeit
     def gen_tile_ann(self) -> List[Dict]:
@@ -125,12 +106,16 @@ def gen_tile_ann(self) -> List[Dict]:
                         coordinates relative to the original image.
         """
         tiles = []
-        pbar = tqdm(total=len(self.dataset))
+        cache_result = []
+        for result in tqdm(self.dataset, desc="Loading dataset annotations..."):
+            cache_result.append(result)
 
-        for idx, result in enumerate(self.dataset):
+        pbar = tqdm(total=len(self.dataset) * 2, desc="Generating tile annotations...")
+        for idx, result in enumerate(cache_result):
             tiles.append(self.gen_single_img(result, dataset_idx=idx))
+            pbar.update(1)
 
-        for idx, result in enumerate(self.dataset):
+        for idx, result in enumerate(cache_result):
             tiles.extend(self.gen_tiles_single_img(result, dataset_idx=idx))
             pbar.update(1)
         return tiles
@@ -145,7 +130,7 @@ def gen_single_img(self, result: Dict, dataset_idx: int) -> Dict:
         Returns:
             Dict: annotation with some other useful information for data pipeline.
         """
-        result["tile_box"] = (0, 0, result["dataset_item"].width, result["dataset_item"].height)
+        result["tile_box"] = (0, 0, result["img_shape"][1], result["img_shape"][0])
         result["dataset_idx"] = dataset_idx
         result["original_shape_"] = result["img_shape"]
         result["uuid"] = str(uuid.uuid4())
@@ -163,11 +148,11 @@ def gen_tiles_single_img(self, result: Dict, dataset_idx: int) -> List[Dict]:
             List[Dict]: a list of tile annotation with some other useful information for data pipeline.
         """
         tile_list = []
-        gt_bboxes = result.pop("gt_bboxes", np.zeros((0, 4), dtype=np.float32))
-        gt_masks = result.pop("gt_masks", None)
-        gt_bboxes_ignore = result.pop("gt_bboxes_ignore", np.zeros((0, 4), dtype=np.float32))
-        gt_labels = result.pop("gt_labels", np.array([], dtype=np.int64))
-        img_shape = result.pop("img_shape")
+        gt_bboxes = result.get("gt_bboxes", np.zeros((0, 4), dtype=np.float32))
+        gt_masks = result.get("gt_masks", None)
+        gt_bboxes_ignore = result.get("gt_bboxes_ignore", np.zeros((0, 4), dtype=np.float32))
+        gt_labels = result.get("gt_labels", np.array([], dtype=np.int64))
+        img_shape = result.get("img_shape")
         height, width = img_shape[:2]
         _tile = self.prepare_result(result)
 
@@ -380,9 +365,15 @@ def __getitem__(self, idx):
         dataset_idx = result["dataset_idx"]
         x_1, y_1, x_2, y_2 = result["tile_box"]
         ori_img = self.dataset[dataset_idx]["img"]
+        cropped_tile = ori_img[y_1:y_2, x_1:x_2, :]
+        tile_path = osp.join(
+            self.tmp_folder, "_".join([str(dataset_idx), result["uuid"], result["ori_filename"], ".jpg"])
+        )
+        self.tiles[idx]["tile_path"] = tile_path
+        mmcv.imwrite(cropped_tile, tile_path)
         if self.img2fp32:
-            ori_img = ori_img.astype(np.float32)
-        result["img"] = ori_img[y_1:y_2, x_1:x_2, :]
+            cropped_tile = cropped_tile.astype(np.float32)
+        result["img"] = cropped_tile
         return result
 
     @staticmethod

otx/algorithms/detection/adapters/openvino/task.py

@@ -84,9 +84,9 @@
     IOptimizationTask,
     OptimizationType,
 )
-from otx.api.utils import Tiler
 from otx.api.utils.dataset_utils import add_saliency_maps_to_dataset_item
 from otx.api.utils.detection_utils import detection2array
+from otx.api.utils.tiler import Tiler
 
 logger = get_logger()
 

otx/api/usecases/exportable_code/demo/demo_package/model_container.py

@@ -15,7 +15,7 @@
 from otx.api.entities.label_schema import LabelSchemaEntity
 from otx.api.entities.model_template import TaskType
 from otx.api.serialization.label_mapper import LabelSchemaMapper
-from otx.api.utils import Tiler
+from otx.api.utils.tiler import Tiler
 from otx.api.utils.detection_utils import detection2array
 
 from .utils import get_model_path, get_parameters

otx/api/usecases/exportable_code/demo/requirements.txt

@@ -1,4 +1,4 @@
 openvino==2022.3.0
 openmodelzoo-modelapi==2022.3.0
-otx @ git+https://github.com/openvinotoolkit/training_extensions/@8c11c3d42c726e6e0eda7364f00cf8ed4dbdc2e9#egg=otx
+otx @ git+https://github.com/openvinotoolkit/training_extensions/@861157a27b5e65f0cde710ee332d4aa55b2ddf45#egg=otx
 numpy>=1.21.0,<=1.23.5  # np.bool was removed in 1.24.0 which was used in openvino runtime

otx/api/utils/__init__.py

@@ -3,8 +3,3 @@
 # Copyright (C) 2021-2022 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 #
-
-from .async_pipeline import OTXDetectionAsyncPipeline
-from .tiler import Tiler
-
-__all__ = ["Tiler", "OTXDetectionAsyncPipeline"]

otx/cli/utils/parser.py

@@ -143,6 +143,11 @@ def _get_leaf_node(curr_dict: Dict[str, dict], curr_key: str):
         value_type = None
         if type_hint is not None:
             value_type = type_hint.get(origin_key, {}).get("type", None)
+        # FIXME[HARIM]: There's no template in args, and it's not inside the workspace, but with --workspace,
+        # the template is not found in args, so params, which are all bools, go into str.
+        # This is a temporary solution.
+        if isinstance(value, str) and value.lower() in ("true", "false"):
+            value_type = str2bool
 
         leaf_node_dict, node_key = _get_leaf_node(params_dict, origin_key)
         leaf_node_dict[node_key] = {"value": value_type(value) if value_type else value}
@@ -249,6 +254,8 @@ def get_parser_and_hprams_data():
         template_config = parse_model_template("./template.yaml")
         hyper_parameters = template_config.hyper_parameters.data
         parser.add_argument("template", nargs="?", default="./template.yaml", help=template_help_str)
+    # TODO: Need fix for how to get hyper_parameters when no template is given and ./template.yaml doesn't exist
+    # Ex. When using --workspace outside of a workspace, but cannot access --workspace from this function.
     else:
         parser.add_argument("template", nargs="?", default=None, help=template_help_str)
 

tests/regression/action/test_action_classification.py

@@ -19,16 +19,21 @@
 )
 from tests.test_suite.e2e_test_system import e2e_pytest_component
 from tests.test_suite.run_test_command import (
-    otx_eval_compare,
-    otx_eval_e2e_eval_time,
-    otx_eval_e2e_train_time,
-    otx_eval_openvino_testing,
     otx_export_testing,
     otx_train_testing,
-    pot_eval_testing,
     pot_optimize_testing,
 )
 
+from tests.regression.regression_command import (
+    regression_eval_testing,
+    regression_openvino_testing,
+    regression_deployment_testing,
+    regression_nncf_eval_testing,
+    regression_pot_eval_testing,
+    regression_train_time_testing,
+    regression_eval_time_testing,
+)
+
 # Configurations for regression test.
 TASK_TYPE = "action_classification"
 TRAIN_TYPE = "supervised"
@@ -68,7 +73,7 @@ def test_otx_train(self, template, tmp_dir_path):
         train_elapsed_time = timer() - train_start_time
 
         infer_start_time = timer()
-        otx_eval_compare(
+        test_result = regression_eval_testing(
             template,
             tmp_dir_path,
             otx_dir,
@@ -82,24 +87,29 @@ def test_otx_train(self, template, tmp_dir_path):
         self.performance[template.name][TIME_LOG["infer_time"]] = round(infer_elapsed_time, 3)
         result_dict[TASK_TYPE][LABEL_TYPE][TRAIN_TYPE]["train"].append(self.performance)
 
+        assert test_result["passed"] is True, test_result["log"]
+
     @e2e_pytest_component
     @pytest.mark.parametrize("template", templates, ids=templates_ids)
     def test_otx_train_kpi_test(self, template):
         results = result_dict[TASK_TYPE][self.label_type][TRAIN_TYPE]["train"]
         performance = get_template_performance(results, template)
 
-        otx_eval_e2e_train_time(
+        kpi_train_result = regression_train_time_testing(
             train_time_criteria=action_cls_regression_config["kpi_e2e_train_time_criteria"]["train"],
             e2e_train_time=performance[template.name][TIME_LOG["train_time"]],
             template=template,
         )
 
-        otx_eval_e2e_eval_time(
+        kpi_eval_result = regression_eval_time_testing(
             eval_time_criteria=action_cls_regression_config["kpi_e2e_eval_time_criteria"]["train"],
             e2e_eval_time=performance[template.name][TIME_LOG["infer_time"]],
             template=template,
         )
 
+        assert kpi_train_result["passed"] is True, kpi_train_result["log"]
+        assert kpi_eval_result["passed"] is True, kpi_eval_result["log"]
+
     @e2e_pytest_component
     @pytest.mark.parametrize("template", templates, ids=templates_ids)
     def test_otx_export_eval_openvino(self, template, tmp_dir_path):
@@ -113,7 +123,7 @@ def test_otx_export_eval_openvino(self, template, tmp_dir_path):
         export_elapsed_time = timer() - export_start_time
 
         export_eval_start_time = timer()
-        otx_eval_openvino_testing(
+        test_result = regression_openvino_testing(
             template,
             tmp_dir_path,
             otx_dir,
@@ -129,6 +139,8 @@ def test_otx_export_eval_openvino(self, template, tmp_dir_path):
         self.performance[template.name][TIME_LOG["export_eval_time"]] = round(export_eval_elapsed_time, 3)
         result_dict[TASK_TYPE][self.label_type][TRAIN_TYPE]["export"].append(self.performance)
 
+        assert test_result["passed"] is True, test_result["log"]
+
     @e2e_pytest_component
     @pytest.mark.parametrize("template", templates, ids=templates_ids)
     def test_pot_optimize_eval(self, template, tmp_dir_path):
@@ -140,7 +152,7 @@ def test_pot_optimize_eval(self, template, tmp_dir_path):
         pot_elapsed_time = timer() - pot_start_time
 
         pot_eval_start_time = timer()
-        pot_eval_testing(
+        test_result = regression_pot_eval_testing(
             template,
             tmp_dir_path,
             otx_dir,
@@ -154,3 +166,5 @@ def test_pot_optimize_eval(self, template, tmp_dir_path):
         self.performance[template.name][TIME_LOG["pot_time"]] = round(pot_elapsed_time, 3)
         self.performance[template.name][TIME_LOG["pot_eval_time"]] = round(pot_eval_elapsed_time, 3)
         result_dict[TASK_TYPE][self.label_type][TRAIN_TYPE]["pot"].append(self.performance)
+
+        assert test_result["passed"] is True, test_result["log"]

tests/regression/action/test_action_detection.py

@@ -19,12 +19,19 @@
 )
 from tests.test_suite.e2e_test_system import e2e_pytest_component
 from tests.test_suite.run_test_command import (
-    otx_eval_compare,
-    otx_eval_e2e_eval_time,
-    otx_eval_e2e_train_time,
     otx_train_testing,
 )
 
+from tests.regression.regression_command import (
+    regression_eval_testing,
+    regression_openvino_testing,
+    regression_deployment_testing,
+    regression_nncf_eval_testing,
+    regression_pot_eval_testing,
+    regression_train_time_testing,
+    regression_eval_time_testing,
+)
+
 # Configurations for regression test.
 TASK_TYPE = "action_detection"
 TRAIN_TYPE = "supervised"
@@ -64,7 +71,7 @@ def test_otx_train(self, template, tmp_dir_path):
         train_elapsed_time = timer() - train_start_time
 
         infer_start_time = timer()
-        otx_eval_compare(
+        test_result = regression_eval_testing(
             template,
             tmp_dir_path,
             otx_dir,
@@ -78,20 +85,25 @@ def test_otx_train(self, template, tmp_dir_path):
         self.performance[template.name][TIME_LOG["infer_time"]] = round(infer_elapsed_time, 3)
         result_dict[TASK_TYPE][LABEL_TYPE][TRAIN_TYPE]["train"].append(self.performance)
 
+        assert test_result["passed"] is True, test_result["log"]
+
     @e2e_pytest_component
     @pytest.mark.parametrize("template", templates, ids=templates_ids)
     def test_otx_train_kpi_test(self, template):
         results = result_dict[TASK_TYPE][self.label_type][TRAIN_TYPE]["train"]
         performance = get_template_performance(results, template)
 
-        otx_eval_e2e_train_time(
+        kpi_train_result = regression_train_time_testing(
             train_time_criteria=action_det_regression_config["kpi_e2e_train_time_criteria"]["train"],
             e2e_train_time=performance[template.name][TIME_LOG["train_time"]],
             template=template,
         )
 
-        otx_eval_e2e_eval_time(
+        kpi_eval_result = regression_eval_time_testing(
             eval_time_criteria=action_det_regression_config["kpi_e2e_eval_time_criteria"]["train"],
             e2e_eval_time=performance[template.name][TIME_LOG["infer_time"]],
             template=template,
         )
+
+        assert kpi_train_result["passed"] is True, kpi_train_result["log"]
+        assert kpi_eval_result["passed"] is True, kpi_eval_result["log"]