Improve Vitens Kiemgetal OTX 1.6.x (#3671)

* modify tile recipes

* remove max_number_det from tile adaptor

* revert change in eval.py

* change default

* udpate description

* udpate description

* change recipe

* update

Author: eugene123tw

src/otx/algorithms/common/configs/training_base.py

@@ -409,7 +409,8 @@ class BaseTilingParameters(ParameterGroup):
 
         tile_max_number = configurable_integer(
             header="Max object per image",
-            description="Max object per image",
+            description="Maximum number of objects per tile. If set to 1500, the tile adaptor "
+            "will automatically determine the value. Otherwise, the manually set value will be used.",
             default_value=1500,
             min_value=1,
             max_value=5000,

src/otx/algorithms/detection/configs/base/data/tiling/atss_tile_pipeline.py

@@ -14,15 +14,9 @@
 img_norm_cfg = dict(mean=[0, 0, 0], std=[255, 255, 255], to_rgb=True)
 
 train_pipeline = [
-    dict(type="MinIoURandomCrop", min_ious=(0.1, 0.3, 0.5, 0.7, 0.9), min_crop_size=0.3),
-    dict(
-        type="Resize",
-        img_scale=[(992, 736), (896, 736), (1088, 736), (992, 672), (992, 800)],
-        multiscale_mode="value",
-        keep_ratio=False,
-    ),
-    dict(type="RandomFlip", flip_ratio=0.5),
+    dict(type="Resize", img_scale=img_size, keep_ratio=False),
     dict(type="Normalize", **img_norm_cfg),
+    dict(type="RandomFlip", flip_ratio=0.5),
     dict(type="DefaultFormatBundle"),
     dict(
         type="Collect",

src/otx/algorithms/detection/configs/base/data/tiling/yolox_tile_pipeline.py

@@ -16,12 +16,6 @@
 img_norm_cfg = dict(mean=[0.0, 0.0, 0.0], std=[1.0, 1.0, 1.0], to_rgb=False)
 
 train_pipeline = [
-    dict(
-        type="RandomAffine",
-        scaling_ratio_range=(0.1, 2),
-        border=(-img_scale[0] // 2, -img_scale[1] // 2),
-    ),
-    dict(type="YOLOXHSVRandomAug"),
     dict(type="RandomFlip", flip_ratio=0.5),
     dict(type="Resize", img_scale=img_scale, keep_ratio=True),
     dict(type="Pad", pad_to_square=True, pad_val=dict(img=(114.0, 114.0, 114.0))),

src/otx/algorithms/detection/configs/detection/configuration.yaml

@@ -645,7 +645,7 @@ tiling_parameters:
 
   tile_max_number:
     header: Max object per tile
-    description: Maximum number of objects per tile
+    description: Maximum number of objects per tile. If set to 1500, the tile adaptor will automatically determine the value. Otherwise, the manually set value will be used.
     affects_outcome_of: TRAINING
     default_value: 1500
     min_value: 1

src/otx/algorithms/detection/configs/detection/cspdarknet_yolox_tiny/tile_pipeline.py

@@ -29,7 +29,6 @@
 img_norm_cfg = dict(mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
 
 train_pipeline = [
-    dict(type="RandomAffine", scaling_ratio_range=(0.5, 1.5), border=(-img_scale[0] // 2, -img_scale[1] // 2)),
     dict(
         type="PhotoMetricDistortion",
         brightness_delta=32,
@@ -62,12 +61,12 @@
 test_pipeline = [
     dict(
         type="MultiScaleFlipAug",
-        img_scale=(416, 416),
+        img_scale=img_scale,
         flip=False,
         transforms=[
             dict(type="Resize", keep_ratio=False),
             dict(type="RandomFlip"),
-            dict(type="Pad", size=(416, 416), pad_val=114.0),
+            dict(type="Pad", size=img_scale, pad_val=114.0),
             dict(type="Normalize", **img_norm_cfg),
             dict(type="ImageToTensor", keys=["img"]),
             dict(type="Collect", keys=["img"]),

src/otx/algorithms/detection/configs/instance_segmentation/configuration.yaml

@@ -646,7 +646,7 @@ tiling_parameters:
 
   tile_max_number:
     header: Max object per tile
-    description: Maximum number of objects per tile
+    description: Maximum number of objects per tile. If set to 1500, the tile adaptor will automatically determine the value. Otherwise, the manually set value will be used.
     affects_outcome_of: TRAINING
     default_value: 1500
     min_value: 1

src/otx/algorithms/detection/configs/rotated_detection/configuration.yaml

@@ -631,7 +631,7 @@ tiling_parameters:
 
   tile_max_number:
     header: Max object per tile
-    description: Maximum number of objects per tile
+    description: Maximum number of objects per tile. If set to 1500, the tile adaptor will automatically determine the value. Otherwise, the manually set value will be used.
     affects_outcome_of: TRAINING
     default_value: 1500
     min_value: 1

src/otx/algorithms/detection/utils/data.py

@@ -490,5 +490,12 @@ def adaptive_tile_params(
     )
 
     tiling_parameters.tile_size = tile_size
-    tiling_parameters.tile_max_number = max_num_objects
     tiling_parameters.tile_overlap = tile_overlap
+
+    if tiling_parameters.tile_max_number == 1500:
+        logger.info(
+            f"----> tile max number is set to default (1500), replace with adaptive max_num_objects: {max_num_objects}"
+        )
+        tiling_parameters.tile_max_number = max_num_objects
+    else:
+        logger.info(f"----> tile max number is manually set to: {tiling_parameters.tile_max_number}")

tests/unit/algorithms/detection/tiling/test_tiling_detection.py

@@ -432,12 +432,15 @@ def test_max_annotation(self, max_annotation=200):
             assert len(data["gt_masks"].data[0][0]) <= max_annotation
 
     @e2e_pytest_unit
-    def test_adaptive_tile_parameters(self):
+    @pytest.mark.parametrize("max_num_img", [100, 300, 1500, 1501])
+    def test_adaptive_tile_parameters(self, max_num_img):
         model_template = parse_model_template(os.path.join(DEFAULT_ISEG_TEMPLATE_DIR, "template.yaml"))
         hp = create(model_template.hyper_parameters.data)
 
         default_tile_size = hp.tiling_parameters.tile_size
         default_tile_overlap = hp.tiling_parameters.tile_overlap
+        # manually set tile max number
+        hp.tiling_parameters.tile_max_number = max_num_img
         default_tile_max_number = hp.tiling_parameters.tile_max_number
 
         adaptive_tile_params(hp.tiling_parameters, self.otx_dataset)
@@ -448,5 +451,9 @@ def test_adaptive_tile_parameters(self):
         # check tile overlap is changed
         assert hp.tiling_parameters.tile_overlap != default_tile_overlap
 
-        # check max output prediction size is changed
-        assert hp.tiling_parameters.tile_max_number != default_tile_max_number
+        if default_tile_max_number == 1500:
+            # check tile max number is being set by adaptive_tile_params
+            assert hp.tiling_parameters.tile_max_number != default_tile_max_number
+        else:
+            # check tile max number is being manually set
+            assert hp.tiling_parameters.tile_max_number == default_tile_max_number