565 feature geo inference integration

config/augmentation/basic_augmentation_segmentation.yaml

@@ -11,7 +11,7 @@ augmentation:
 
   # Augmentations done immediately before conversion to torch tensor
   normalization: # Normalization: parameters for finetuning. See examples below:
-    mean:  # -> mean: [0.485, 0.456, 0.406]
-    std:  # -> std: std: [0.229, 0.224, 0.225])
+    mean: [1e-8, 1e-8, 1e-8] # -> mean: [0.485, 0.456, 0.406]
+    std: [1e-8, 1e-8, 1e-8] # -> std: std: [0.229, 0.224, 0.225])
   scale_data: [ 0, 1 ]  # Min and max value to scale values of input imagery
 

config/inference/default_binary.yaml

@@ -3,13 +3,13 @@ inference:
   raw_data_csv: tests/inference/inference_segmentation_binary.csv
   root_dir: inferences
   input_stac_item:  # alternatively, use a path or url to stac item directly
-  state_dict_path: ${general.save_weights_dir}/
+  model_path: ${general.save_weights_dir}/
   output_path:
   checkpoint_dir:  # (string, optional): directory in which to save the object if url
+  batch_size: 8
   chunk_size:  # if empty, will be calculated automatically from max_pix_per_mb_gpu
   # Maximum number of pixels each Mb of GPU Ram to allow. E.g. if GPU has 1000 Mb of Ram and this parameter is set to
   # 10, chunk_size will be set to sqrt(1000 * 10) = 100.
-  use_hanning: True # enables smoothening with Hann windows; creates overlapping tiles.
   max_pix_per_mb_gpu: 25
   prep_data_only: False
   override_model_params: False
@@ -22,4 +22,4 @@ inference:
   max_used_ram: ${training.max_used_ram}  # If RAM usage of detected GPU exceeds this percentage, it will be ignored
 
   # Post-processing
-  ras2vec: False  # if True, a polygonized version of the inference (.gpkg) will be created with rasterio tools
+  mask_to_vector: False  # if True, a polygonized version of the inference (.gpkg) will be created with rasterio tools

config/inference/default_multiclass.yaml

@@ -3,9 +3,10 @@ inference:
   raw_data_csv: tests/inference/inference_segmentation_multiclass.csv
   root_dir: inferences
   input_stac_item:  # alternatively, use a path or url to stac item directly
-  state_dict_path: ${general.save_weights_dir}/
+  model_path: ${general.save_weights_dir}/
   output_path:
   checkpoint_dir:  # (string, optional): directory in which to save the object if url
+  batch_size: 8
   chunk_size:  # if empty, will be calculated automatically from max_pix_per_mb_gpu
   # Maximum number of pixels each Mb of GPU Ram to allow. E.g. if GPU has 1000 Mb of Ram and this parameter is set to
   # 10, chunk_size will be set to sqrt(1000 * 10) = 100.
@@ -21,4 +22,4 @@ inference:
   max_used_ram: ${training.max_used_ram}  # If RAM usage of detected GPU exceeds this percentage, it will be ignored
 
   # Post-processing
-  ras2vec: False  # if True, a polygonized version of the inference (.gpkg) will be created with rasterio tools
+  mask_to_vector: False  # if True, a polygonized version of the inference (.gpkg) will be created with rasterio tools

config/training/default_training.yaml

@@ -13,7 +13,7 @@ training:
   max_used_perc:
   state_dict_path:
   state_dict_strict_load: True
-  script_model: False
+  script_model: True
   compute_sampler_weights: False
 
 #  precision: 16

environment.yml

@@ -1,6 +1,7 @@
 name: geo_deep_env
 channels:
   - pytorch
+  - nvidia
   - conda-forge
 dependencies:
   - coverage>=6.3.1
@@ -18,10 +19,11 @@ dependencies:
   - ruamel_yaml>=0.15
   - scikit-image>=0.18
   - torchgeo>=0.3
-  - torchvision>=0.12
+  - torchvision>=0.13
   - pip:
+    - geo-inference>=2.0.5
     - hydra-colorlog>=1.1.0
     - hydra-optuna-sweeper>=1.1.0
+    - segmentation-models-pytorch>=0.3.3
     - git+https://github.com/CosmiQ/solaris.git@0.5.0
-    - ttach>=0.0.3
     - mlflow>=1.2 # causes env solving to hang if not with pip

evaluate_segmentation.py

@@ -124,7 +124,9 @@ def main(params):
     gpkg_name_ = []
 
     for aoi in tqdm(list_aois, desc='Evaluating from input list', position=0, leave=True):
-        output_path = working_folder / f"{aoi.aoi_id}_pred.tif" if not inference_image else inference_image
+        print(f"\nEvaluating AOI: {aoi.aoi_id}")
+        print(f"working_folder: {working_folder}")
+        output_path = working_folder / f"{aoi.aoi_id}_mask.tif" if not inference_image else inference_image
         if not output_path.is_file():
             raise FileNotFoundError(f"Couldn't locate inference to evaluate metrics with. Make inference has been run "
                                     f"before you run evaluate mode.")