transformers_as_arg

geo_inference/config/sample.yaml

@@ -9,6 +9,9 @@ arguments:
   vec: False                      # Vector Conversion: bool 
   yolo: False                     # YOLO Conversion: bool
   coco: False                     # COCO Conversion: bool
+  transformers : True
+  transformer_flip : False
+  transformer_rotate : True
   device: "gpu"                   # cpu or gpu: str
   gpu_id: 0
   mgpu: False

geo_inference/geo_inference.py

@@ -11,6 +11,7 @@
 import threading
 import numpy as np
 import xarray as xr
+import ttach as tta
 from typing import Dict
 from dask import config
 import dask.array as da
@@ -83,6 +84,9 @@ def __init__(
         gpu_id: int = 0,
         num_classes: int = 5,
         prediction_threshold : float = 0.3,
+        transformers : bool = False,
+        transformer_flip: bool = False,
+        transformer_rotate: bool = False,
     ):
         self.work_dir: Path = get_directory(work_dir)
         self.device = (
@@ -95,6 +99,23 @@ def __init__(
             ),
             map_location=self.device,
         )
+        if transformers:
+            if transformer_flip and transformer_rotate:    # do all
+                transforms = tta.aliases.d4_transform()
+            elif transformer_rotate:                       # do rotate only
+                transforms = tta.Compose(
+                    [
+                        tta.Rotate90(angles=[90]),
+                    ]
+                )
+            elif transformer_flip:                         # do flip only
+                transforms = tta.Compose(
+                    [
+                        tta.HorizontalFlip(),
+                        tta.VerticalFlip(),
+                    ]
+                )
+            self.model = tta.SegmentationTTAWrapper(self.model, transforms, merge_mode='mean')
         self.mask_to_vec = mask_to_vec
         self.mask_to_coco = mask_to_coco
         self.mask_to_yolo = mask_to_yolo
@@ -363,7 +384,10 @@ def main() -> None:
         device=arguments["device"],
         gpu_id=arguments["gpu_id"],
         num_classes=arguments["classes"],
-        prediction_threshold=arguments["prediction_threshold"]
+        prediction_threshold=arguments["prediction_threshold"],
+        transformers=arguments["transformers"],
+        transformer_flip=arguments["transformer_flip"],
+        transformer_rotate=arguments["transformer_rotate"],
     )
     inference_mask_layer_name = geo_inference(
         inference_input=arguments["image"],
@@ -372,6 +396,7 @@ def main() -> None:
         workers=arguments["workers"],
         bbox=arguments["bbox"],
     )
+    print(inference_mask_layer_name)
 
 
 

geo_inference/utils/helpers.py

@@ -433,6 +433,10 @@ def cmd_interface(argv=None):
 
     parser.add_argument("-pr", "--prediction_thr", type=float, nargs=1, help="Prediction Threshold")
 
+    parser.add_argument("-tr", "--transformers", nargs=1, help="Transformers Addition")
+    parser.add_argument("-tr_f", "--transformer_flip", nargs=1, help="Transformers Addition - Flip")
+    parser.add_argument("-tr_e", "--transformer_rotate", nargs=1, help="Transformers Addition - Rotate")
+
     args = parser.parse_args()
 
     if args.args:
@@ -452,6 +456,10 @@ def cmd_interface(argv=None):
         classes = config["arguments"]["classes"]
         patch_size = config["arguments"]["patch_size"]
         prediction_threshold = config["arguments"]["prediction_thr"]
+        transformers = config["arguments"]["transformers"]
+        transformer_flip = config["arguments"]["transformer_flip"]
+        transformer_rotate = config["arguments"]["transformer_rotate"]
+
     elif args.image:
         image =args.image[0]
         model = args.model[0] if args.model else None
@@ -468,6 +476,10 @@ def cmd_interface(argv=None):
         classes = args.classes[0] if args.classes else 5
         patch_size = args.patch_size[0] if args.patch_size else 1024 
         prediction_threshold = args.prediction_thr[0] if args.prediction_thr else 0.3
+        transformers = args.transformers[0] if args.transformers else False
+        transformer_flip = args.transformer_flip if args.transformer_flip else False
+        transformer_rotate = args.transformer_rotate if args.transformer_rotate else False
+
     else:
         print("use the help [-h] option for correct usage")
         raise SystemExit
@@ -487,6 +499,9 @@ def cmd_interface(argv=None):
         "gpu_id": gpu_id,
         "patch_size": patch_size,
         "prediction_threshold": prediction_threshold,
+        "transformers": transformers,
+        "transformer_flip": transformer_flip,
+        "transformer_rotate":transformer_rotate,
     }
     return arguments
 

requirements.txt

@@ -10,4 +10,5 @@ dask[distributed]>=2024.6.2
 requests>=2.32.3
 xarray>=2024.6.0 
 pystac>=1.10.1
-rioxarray>=0.15.6
+rioxarray>=0.15.6
+ttach>=0.0.3

tests/data/sample.yaml

@@ -14,4 +14,7 @@ arguments:
   classes : 5
   n_workers: 20
   prediction_thr : 0.3
+  transformers: False
+  transformer_flip : False
+  transformer_rotate : False
   patch_size: 1024

tests/utils/test_helpers.py

@@ -57,6 +57,9 @@ def test_read_yaml(test_data_dir):
                                     "mgpu": False,
                                     "classes": 5,
                                     "prediction_thr": 0.3,
+                                    "transformers": False,
+                                    "transformer_flip" : False,
+                                    "transformer_rotate" : False,
                                     "n_workers": 20
                                    }
 
@@ -145,6 +148,9 @@ def test_cmd_interface_with_args(monkeypatch, test_data_dir):
                       "classes": 5,
                       "multi_gpu": False,
                       "prediction_threshold": 0.3,
+                      "transformers": False,
+                      "transformer_flip" : False,
+                      "transformer_rotate" : False,
                       "patch_size": 1024
                       }
 
@@ -169,6 +175,9 @@ def test_cmd_interface_with_image(monkeypatch):
         "gpu_id": 0,
         "classes": 5,
         "prediction_threshold": 0.3,
+        "transformers": False,
+        "transformer_flip" : False,
+        "transformer_rotate" : False,
         "multi_gpu": False,
     }