target models

conf/config.yaml

@@ -6,7 +6,8 @@ comet:
   workspace: bw4sz
 
 check_annotations: true
-force_upload: false
+# Force upload bypasses the pipeline, useful for debugging and starting a new project
+force_upload: true
 label_studio:
   project_name: "Bureau of Ocean Energy Management"
   url: "https://labelstudio.naturecast.org/"
@@ -16,7 +17,7 @@ label_studio:
   csv_dir: /blue/ewhite/b.weinstein/BOEM/annotations
 
 predict:
-  patch_size: 450
+  patch_size: 2000
   patch_overlap: 0
   min_score: 0.5
 
@@ -68,26 +69,26 @@ active_learning:
   image_dir: /blue/ewhite/b.weinstein/BOEM/sample_flight/JPG_2024_Jan27/annotated
   strategy: 'target-labels'
   n_images: 10
-  patch_size: 256
-  patch_overlap: 0.5
-  min_score: 0.5
+  patch_size: 2000
+  patch_overlap: 0
+  min_score: 0.2
   model_checkpoint:
   target_labels:
     - "Bird"
 
   # Optional parameters:
   evaluation: 
   dask_client: 
-  pool_limit: 100
+  pool_limit: 500
 
 active_testing:
   image_dir: /blue/ewhite/b.weinstein/BOEM/sample_flight/JPG_2024_Jan27/annotated
   strategy: 'random'
-  n_images: 100
-  m: 10
-  patch_size: 256
-  patch_overlap: 0.5
-  min_score: 0.5
+  n_images: 10
+  m: 
+  patch_size: 2000
+  patch_overlap: 0
+  min_score: 0.2
 
 deepforest:
   train:

src/active_learning.py

@@ -45,8 +45,8 @@ def choose_train_images(evaluation, image_dir, strategy, n=10, patch_size=512, p
         return chosen_images    
     elif strategy in ["most-detections","target-labels"]:
         # Predict all images
-        if model_path is None:
-            raise ValueError("A model is required for the 'most-detections' or 'target-labels' strategy.")
+        if model_path is None and model is None:
+            raise ValueError("A model is required for the 'most-detections' or 'target-labels' strategy. Either pass a model or a model_path.")
         if dask_client:
             # load model on each client
             def update_sys_path():
@@ -59,7 +59,7 @@ def update_sys_path():
             blocks = dask_pool.to_delayed().ravel()
             block_futures = []
             for block in blocks:
-                block_future = dask_client.submit(detection.predict,image_paths=block.compute(), patch_size=patch_size, patch_overlap=patch_overlap, min_score=min_score, model_path=model_path)
+                block_future = dask_client.submit(detection.predict,image_paths=block.compute(), patch_size=patch_size, patch_overlap=patch_overlap, model_path=model_path)
                 block_futures.append(block_future)
             # Get results
             dask_results = []
@@ -68,7 +68,7 @@ def update_sys_path():
                 dask_results.append(pd.concat(block_result))
             preannotations = pd.concat(dask_results)
         else:
-            preannotations = detection.predict(model=model, image_paths=pool, patch_size=patch_size, patch_overlap=patch_overlap, min_score=min_score)
+            preannotations = detection.predict(m=model, image_paths=pool, patch_size=patch_size, patch_overlap=patch_overlap)
             preannotations = pd.concat(preannotations)
 
         if strategy == "most-detections":

src/pipeline.py

@@ -11,7 +11,7 @@
 from src import label_studio
 from src.classification import preprocess_and_train_classification
 from src.data_processing import density_cropping
-from src.detection import preprocess_and_train
+from src.detection import preprocess_and_train, load
 from src.pipeline_evaluation import PipelineEvaluation
 from src.reporting import Reporting
 
@@ -47,16 +47,19 @@ def run(self):
             if new_annotations is None:
                 print("No new annotations, exiting")
                 if self.config.force_upload:
-                    image_paths = glob.glob(os.path.join(self.config.label_studio.images_to_annotate_dir, "*.jpg"))
-                    image_paths = random.sample(image_paths, 10)
-                    label_studio.upload_to_label_studio(images=image_paths, 
-                                                    sftp_client=self.sftp_client, 
-                                                    label_studio_project=self.label_studio_project, 
-                                                    images_to_annotate_dir=self.config.label_studio.images_to_annotate_dir,
-                                                    folder_name=self.config.label_studio.folder_name, 
-                                                        preannotations=None
-                                                           )
-                    return None
+                    detection_model = load(self.config.detection_model.checkpoint)
+                    train_images_to_annotate = choose_train_images(
+                    evaluation=None,
+                    image_dir=self.config.active_learning.image_dir,
+                    model=detection_model,
+                    strategy=self.config.active_learning.strategy,
+                    n=self.config.active_learning.n_images,
+                    patch_size=self.config.active_learning.patch_size,
+                    patch_overlap=self.config.active_learning.patch_overlap,
+                    min_score=self.config.active_learning.min_score,
+                    target_labels=self.config.active_learning.target_labels
+                )
+                return None
                     # Select images to upload
 
                 return None
@@ -106,7 +109,8 @@ def run(self):
                 n=self.config.active_learning.n_images,
                 patch_size=self.config.active_learning.patch_size,
                 patch_overlap=self.config.active_learning.patch_overlap,
-                min_score=self.config.active_learning.min_score
+                min_score=self.config.active_learning.min_score,
+                target_labels=self.config.active_learning.target_labels
             )
 
             test_images_to_annotate = choose_test_images(