Merge pull request #6 from encord-team/fhv/all_image_features

Compute all image features.

pocs/img_features.py

@@ -42,34 +42,46 @@ def rank_by_blur(image):
     return 1 - cv2.Laplacian(image, cv2.CV_64F).var()
 
 
+options = {
+    "contrast": (rank_by_contrast, "Contrast"),
+    "green": (rank_by_green, "Greenness"),
+    "red": (rank_by_red, "Redness"),
+    "blue": (rank_by_blue, "Blueness"),
+    "brightness": (rank_by_brightness, "Brightness"),
+    "sharpness": (rank_by_sharpness, "Sharpness"),
+    "blurriness": (rank_by_blur, "Blurriness"),
+}
+
+
 @hydra.main(version_base=None, config_path="../conf", config_name="config")
 def main(cfg: DictConfig):
     cache_dir, project = fetch_project_info(cfg)
-    iterator = DatasetIterator(project, cache_dir, use_images=True, subset_size=100)
+    iterator = DatasetIterator(project, cache_dir, use_images=True, subset_size=-1)
 
     # When you log stuff, this will be stored in the `pocs/outputs` directory.
 
     # Preprocessing happens here.
     # You can build/load databases of embeddings, compute statistics etc,
     # ...
 
-    with CSVIndexWriter(cache_dir, iterator, prefix="brightness-indexer") as writer:
-        for data_unit, img_pth in tqdm(iterator.iterate(), desc="Iterating data units", total=len(iterator)):
-            key = iterator.get_identifier()
+    for rank_fn, name in options.values():
+        with CSVIndexWriter(cache_dir, iterator, prefix=f"{name}-indexer") as writer:
+            for data_unit, img_pth in tqdm(iterator.iterate(), desc=f"Looking for {name}", total=len(iterator)):
+                key = iterator.get_identifier()
 
-            # This is where you do the actual inference.
+                # This is where you do the actual inference.
 
-            # Some convenient properties of the current data.
-            # ``iterator.label_hash`` the label hash of the current data unit
-            # ``iterator.du_hash`` the data unit hash of
-            # ``iterator.frame`` the frame of the current data unit hash of
-            # ``iterator.num_frame`` the total number of frames in the label row.
+                # Some convenient properties of the current data.
+                # ``iterator.label_hash`` the label hash of the current data unit
+                # ``iterator.du_hash`` the data unit hash of
+                # ``iterator.frame`` the frame of the current data unit hash of
+                # ``iterator.num_frame`` the total number of frames in the label row.
 
-            # Do your thing (inference)
-            image = cv2.imread(img_pth.as_posix())
-            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+                # Do your thing (inference)
+                image = cv2.imread(img_pth.as_posix())
+                image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
 
-            writer.write_score(key, rank_by_brightness(image), "Brightness")
+                writer.write_score(key, rank_fn(image), "")
 
 
 if __name__ == '__main__':