Adaptive feature quality

opendm/osfm.py

@@ -13,7 +13,7 @@ from opendm import system
 from opendm import context
 from opendm import camera
 from opendm import location
-from opendm.photo import find_largest_photo_dim, find_largest_photo
+from opendm.photo import find_largest_photo_dims, find_largest_photo
 from opensfm.large import metadataset
 from opensfm.large import tools
 from opensfm.actions import undistort
@@ -211,11 +211,25 @@ class OSFMContext:
                 'lowest': 0.0675,
             }
 
-            max_dim = find_largest_photo_dim(photos)
+            max_dims = find_largest_photo_dims(photos)
 
-            if max_dim > 0:
+            if max_dims is not None:
+                w, h = max_dims
+                max_dim = max(w, h)
                 log.ODM_INFO("Maximum photo dimensions: %spx" % str(max_dim))
-                feature_process_size = int(max_dim * feature_quality_scale[args.feature_quality])
+
+                lower_limit = 320
+                upper_limit = 4480
+                megapixels = (w * h) / 1e6
+                multiplier = 1
+                    
+                if megapixels < 2:
+                    multiplier = 2
+                elif megapixels > 42:
+                    multiplier = 0.5
+                
+                factor = min(1, feature_quality_scale[args.feature_quality] * multiplier)
+                feature_process_size = min(upper_limit, max(lower_limit, int(max_dim * factor)))
                 log.ODM_INFO("Photo dimensions for feature extraction: %ipx" % feature_process_size)
             else:
                 log.ODM_WARNING("Cannot compute max image dimensions, going with defaults")
@@ -283,9 +297,8 @@ class OSFMContext:
             config.append("matcher_type: %s" % osfm_matchers[matcher_type])
 
             # GPU acceleration?
-            if has_gpu(args):
-                max_photo = find_largest_photo(photos)
-                w, h = max_photo.width, max_photo.height
+            if has_gpu(args) and max_dims is not None:
+                w, h = max_dims
                 if w > h:
                     h = int((h / w) * feature_process_size)
                     w = int(feature_process_size)