use only first image to calculate chunk size

cvat/apps/engine/media_extractors.py

@@ -66,7 +66,7 @@ def image_names(self):
         pass
 
     @abstractmethod
-    def get_avg_image_size(self, percent_to_analize=10):
+    def get_avg_image_size(self):
         pass
 
 #Note step, start, stop have no affect
@@ -97,15 +97,9 @@ def save_preview(self, preview_path):
     def image_names(self):
         return self._source_path
 
-    def get_avg_image_size(self, percent_to_analize=10):
-        widths = []
-        heights = []
-        step = 100 // percent_to_analize or 1
-        for img_path in itertools.islice(self._source_path, 0, None, step):
-            img = Image.open(img_path)
-            widths.append(img.width)
-            heights.append(img.height)
-        return (sum(widths) // len(widths), sum(heights) // len(heights))
+    def get_avg_image_size(self):
+        img = Image.open(self._source_path[0])
+        return img.width, img.height
 
 #Note step, start, stop have no affect
 class DirectoryReader(ImageListReader):
@@ -197,16 +191,9 @@ def save_preview(self, preview_path):
         with open(preview_path, 'wb') as f:
             f.write(self._zip_source.read(self._source_path[0]))
 
-    def get_avg_image_size(self, percent_to_analize=10):
-        widths = []
-        heights = []
-        step = 100 // percent_to_analize or 1
-        for img_path in itertools.islice(self._source_path, 0, None, step):
-            img = Image.open(BytesIO(self._zip_source.read(img_path)))
-            widths.append(img.width)
-            heights.append(img.height)
-        return (sum(widths) // len(widths), sum(heights) // len(heights))
-
+    def get_avg_image_size(self):
+        img = Image.open(BytesIO(self._zip_source.read(self._source_path[0])))
+        return img.width, img.height
 
     @property
     def image_names(self):
@@ -259,9 +246,9 @@ def save_preview(self, preview_path):
     def image_names(self):
         return self._source_path
 
-    def get_avg_image_size(self, percent_to_analize=10):
-        first_image = (next(iter(self)))[0]
-        return first_image.width, first_image.height
+    def get_avg_image_size(self):
+        image = (next(iter(self)))[0]
+        return image.width, image.height
 
 class IChunkWriter(ABC):
     def __init__(self, quality):

cvat/apps/engine/task.py

@@ -254,19 +254,22 @@ def update_progress(progress):
     original_chunk_writer = original_chunk_writer_class(100)
 
     # calculate chunk size if it isn't specified
-    if db_data.chunk_size is None and isinstance(compressed_chunk_writer, ZipCompressedChunkWriter):
-        avg_w, avg_h = extractor.get_avg_image_size()
-        avg_area = avg_h * avg_w
-        if avg_area <= 1920 * 1080:
-            db_data.chunk_size = 36
-        elif avg_area <= 2560 * 1440:
-            db_data.chunk_size = 18
-        elif avg_area <= 3840 * 2160:
-            db_data.chunk_size = 9
-        elif avg_area <= 5120 * 2880:
-            db_data.chunk_size = 4
+    if db_data.chunk_size is None:
+        if isinstance(compressed_chunk_writer, ZipCompressedChunkWriter):
+            avg_w, avg_h = extractor.get_avg_image_size()
+            avg_area = avg_h * avg_w
+            if avg_area <= 1920 * 1080:
+                db_data.chunk_size = 36
+            elif avg_area <= 2560 * 1440:
+                db_data.chunk_size = 18
+            elif avg_area <= 3840 * 2160:
+                db_data.chunk_size = 9
+            elif avg_area <= 5120 * 2880:
+                db_data.chunk_size = 4
+            else:
+                db_data.chunk_size = 2
         else:
-            db_data.chunk_size = 2
+            db_data.chunk_size = 36
 
     frame_counter = 0
     total_len = len(extractor) or 100