changes based on comments

Author: yawenzzzz

docker-compose.yaml

@@ -3,34 +3,3 @@ version: "3.9"
 services:
   test:
     build: .
-
-  # Define the base image
-  base-image:
-    build:
-      context: .
-      dockerfile: Dockerfile
-    image: base-image:latest   # Tag it as "base-image"
-
-  # Define the landsat-vessels service
-  landsat-vessels:
-    build:
-      context: ./rslp/landsat_vessels
-      dockerfile: Dockerfile
-    shm_size: '10G'  # This adds the shared memory size
-    depends_on:
-      - base-image
-    ports:
-      - "5555:5555"
-    environment:
-      - RSLP_BUCKET
-      - S3_ACCESS_KEY_ID
-      - S3_SECRET_ACCESS_KEY
-      - AWS_ACCESS_KEY_ID
-      - AWS_SECRET_ACCESS_KEY
-      - NVIDIA_VISIBLE_DEVICES=all  # Make all GPUs visible
-    deploy:
-      resources:
-        reservations:
-          devices:
-            - capabilities: [gpu]  # Ensure this service can access GPUs
-    runtime: nvidia  # Use the NVIDIA runtime

landsat/recheck_landsat_labels/phase123_config.yaml

@@ -22,10 +22,10 @@ model:
     plateau_patience: 10
     plateau_min_lr: 0
     plateau_cooldown: 0
-    restore_config:
-      restore_path: gcs://rslearn-eai/datasets/landsat_vessel_detection/artifacts/2024-03-13-landsat-vessels/vessel02_satlas_freeze_crop512_nosatlas_b8first2/best.pth
-      remap_prefixes:
-        - ["backbone.backbone.", "encoder.0.model."]
+    # restore_config:
+    #   restore_path: gcs://rslearn-eai/datasets/landsat_vessel_detection/artifacts/2024-03-13-landsat-vessels/vessel02_satlas_freeze_crop512_nosatlas_b8first2/best.pth
+    #   remap_prefixes:
+    #     - ["backbone.backbone.", "encoder.0.model."]
 data:
   class_path: rslearn.train.data_module.RslearnDataModule
   init_args:
@@ -52,6 +52,8 @@ data:
               allow_invalid: true
               skip_unknown_categories: true
               prob_property: "prob"
+              positive_class: "correct"
+              positive_class_threshold: 0.85
         input_mapping:
           class:
             label: "targets"

rslp/landsat_vessels/docker-compose.yaml

@@ -0,0 +1,33 @@
+version: "3.9"
+
+services:
+  # Define the base image
+  base-image:
+    build:
+      context: ../..
+      dockerfile: Dockerfile
+    image: base-image:latest   # Tag it as "base-image"
+
+  # Define the landsat-vessels service
+  landsat-vessels:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    shm_size: '10G'  # This adds the shared memory size
+    depends_on:
+      - base-image
+    ports:
+      - "5555:5555"
+    environment:
+      - RSLP_BUCKET
+      - S3_ACCESS_KEY_ID
+      - S3_SECRET_ACCESS_KEY
+      - AWS_ACCESS_KEY_ID
+      - AWS_SECRET_ACCESS_KEY
+      - NVIDIA_VISIBLE_DEVICES=all  # Make all GPUs visible
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - capabilities: [gpu]  # Ensure this service can access GPUs
+    runtime: nvidia  # Use the NVIDIA runtime

rslp/landsat_vessels/predict_pipeline.py

@@ -205,6 +205,7 @@ def predict_pipeline(
             specified.
     """
     start_time = time.time()  # Start the timer
+    time_profile = {}
 
     ds_path = UPath(scratch_path)
     ds_path.mkdir(parents=True, exist_ok=True)
@@ -264,15 +265,23 @@ def predict_pipeline(
             dst_geom.time_range[1] + timedelta(minutes=30),
         )
 
+    time_profile["setup"] = time.time() - start_time
+
     # Run pipeline.
-    print("get vessel detections")
+    step_start_time = time.time()
+    print("run detector")
     detections = get_vessel_detections(
         ds_path, projection, scene_bounds, time_range=time_range
     )
+    time_profile["get_vessel_detections"] = time.time() - step_start_time
+
+    step_start_time = time.time()
     print("run classifier")
     detections = run_classifier(ds_path, detections, time_range=time_range)
+    time_profile["run_classifier"] = time.time() - step_start_time
 
     # Write JSON and crops.
+    step_start_time = time.time()
     crop_path = UPath(crop_path)
     crop_path.mkdir(parents=True, exist_ok=True)
 
@@ -332,12 +341,18 @@ def predict_pipeline(
             )
         )
 
+    time_profile["write_json_and_crops"] = time.time() - step_start_time
+
     elapsed_time = time.time() - start_time  # Calculate elapsed time
-    print(f"Prediction pipeline completed in {elapsed_time:.2f} seconds")
+    time_profile["total"] = elapsed_time
 
     if json_path:
         json_path = UPath(json_path)
         with json_path.open("w") as f:
             json.dump(json_data, f)
 
+    print(f"Prediction pipeline completed in {elapsed_time:.2f} seconds")
+    for step, duration in time_profile.items():
+        print(f"{step} took {duration:.2f} seconds")
+
     return json_data