Limit processing

python_utils/running_model.py

@@ -121,7 +121,8 @@ def run_predictions(files):
     results = []
 
     ## When using GPU, run single instance
-    if torch.cuda.is_available():
+    ## Or if not checking many files, as loading multiple models isn't worthwhile
+    if torch.cuda.is_available() or len(files) < 4:
         for file in files:
             # Run prediction with original BGR image
             predictor_ = load_model()
@@ -129,14 +130,15 @@ def run_predictions(files):
 
             results.append(prediction_output)
     else:
-        ## Else use multiprocessing to run in parallel
-        print(f'Using {multiprocessing.cpu_count()} cpus')
+        ## Else use multiprocessing to run in parallel with 2 processes
+
+        print(f'Using {multiprocessing.cpu_count()} cpus apparently')
         # Set up to load one model per worker process
         def init_worker():
             global predictor
             predictor = load_model()  # Load once per worker process
 
-        with multiprocessing.Pool(initializer=init_worker) as pool:
+        with multiprocessing.Pool(initializer=init_worker, processes=2) as pool:
             results = pool.map(process_file, files)
     return results