Bug fix for Image2Struct

src/helm/benchmark/metrics/vision_language/emd_utils.py

@@ -280,6 +280,10 @@ def compute_emd_recursive(
     assert max_num_patches > 0
     assert 0 < weight_most_frequent_color <= 1
 
+    # Convert the images to RGB first. Some images have 4 channels (RGBA)
+    img1_PIL = img1_PIL.convert("RGB")
+    img2_PIL = img2_PIL.convert("RGB")
+
     # Resize the images so that there are not too many patches
     # Try to maintain the aspect ratio and resize to a multiple of the patch size
     num_patches = math.ceil(img1_PIL.size[0] / patch_size[0]) * math.ceil(img1_PIL.size[1] / patch_size[1])

src/helm/benchmark/metrics/vision_language/image_metrics.py

@@ -78,9 +78,9 @@ class AnnotatedImageMetrics(Metric):
 
     # Metric names
     COMPILE_METRIC: str = "compilation_success"
-    BLOCK_EARTH_MOVER_SIMILARITY_NORM1: str = "block_emd_similarity_white"
-    BLOCK_EARTH_MOVER_SIMILARITY_NORM2: str = "block_emd_similarity_median_color"
-    BLOCK_EARTH_MOVER_SIMILARITY: str = "block_emd_similarity"
+    EARTH_MOVER_SIMILARITY = "earth_mover_similarity"
+    EARTH_MOVER_SIMILARITY_WHITE = "earth_mover_similarity_white"
+    BLOCK_EMD: str = "block_emd"
     PIXEL_SIMILARITY: str = "pixel_similarity"
     SIFT_SIMILARITY: str = "sift_similarity"
     LPIPS_SIMILARITY: str = "lpips_similarity"
@@ -108,12 +108,13 @@ def __init__(self, generation_type: str, metric_names: List[str], size_handling_
         metrics: List[AnnotatedMetric] = [
             AnnotatedMetric(self.PIXEL_SIMILARITY, pixel_similarity, "image_np_gray"),
             AnnotatedMetric(self.SIFT_SIMILARITY, sift_similarity, "image_np"),
-            # Raw block EMD
-            AnnotatedMetric(self.BLOCK_EARTH_MOVER_SIMILARITY, self.compute_block_emd_raw, "image_PIL"),
-            # Normalized block EMD against white
-            AnnotatedMetric(self.BLOCK_EARTH_MOVER_SIMILARITY_NORM1, self.compute_block_emd_white, "image_PIL"),
-            # Normalized block EMD against median
-            AnnotatedMetric(self.BLOCK_EARTH_MOVER_SIMILARITY_NORM2, self.compute_block_emd_extreme, "image_PIL"),
+            AnnotatedMetric(self.BLOCK_EMD, self.compute_block_emd_raw, "image_PIL"),  # Raw block-EMD
+            AnnotatedMetric(
+                self.EARTH_MOVER_SIMILARITY_WHITE, self.ems_white, "image_PIL"
+            ),  # Normalized block-EMD against white
+            AnnotatedMetric(
+                self.EARTH_MOVER_SIMILARITY, self.ems, "image_PIL"
+            ),  # Normalized block-EMD against black/white
             AnnotatedMetric(self.LPIPS_SIMILARITY, self.lpips_similarity, "image_PIL"),
             AnnotatedMetric(self.FID_SIMILARITY, self.fid_similarity, "image_PIL"),
             AnnotatedMetric(self.SSIM_SIMILARITY, self.compute_ssim, "image_np_gray"),
@@ -414,7 +415,7 @@ def compute_edit_sim(self, completion: str, reference: str) -> float:
         result = _edit_similarity(completion_tokens, truncated_reference_tokens)
         return result
 
-    def compute_block_emd_white(
+    def ems_white(
         self,
         pred_image: Image.Image,
         ref_image: Image.Image,
@@ -455,17 +456,18 @@ def compute_denominator():
 
         hash_dict = {
             "reference_image": str(AnnotatedImageMetrics.HASH_FUNC(ref_image, hash_size=self.HASH_LENGTH)),
+            "generated_image": str(AnnotatedImageMetrics.HASH_FUNC(pred_image, hash_size=self.HASH_LENGTH)),
         }
-        cache_key_numerator = {"metric_name": f"intermediate_{self.BLOCK_EARTH_MOVER_SIMILARITY}", **hash_dict}
-        cache_key_denominator = {"metric_name": f"intermediate_{self.BLOCK_EARTH_MOVER_SIMILARITY_NORM1}", **hash_dict}
+        cache_key_numerator = {"metric_name": f"intermediate_{self.BLOCK_EMD}", **hash_dict}
+        cache_key_denominator = {"metric_name": "intermediate_ems_white_denominator", **hash_dict}
 
         assert self._cache is not None
         emd_raw, _ = self._cache.get(cache_key_numerator, compute_numerator)
         emd_base, _ = self._cache.get(cache_key_denominator, compute_denominator)
 
-        return 1.0 - emd_raw["value"] / emd_base["value"]
+        return max(0, 1.0 - emd_raw["value"] / emd_base["value"])
 
-    def compute_block_emd_extreme(
+    def ems(
         self,
         pred_image: Image.Image,
         ref_image: Image.Image,
@@ -513,9 +515,10 @@ def compute_denominator():
 
         hash_dict = {
             "reference_image": str(AnnotatedImageMetrics.HASH_FUNC(ref_image, hash_size=self.HASH_LENGTH)),
+            "generated_image": str(AnnotatedImageMetrics.HASH_FUNC(pred_image, hash_size=self.HASH_LENGTH)),
         }
-        cache_key_numerator = {"metric_name": f"intermediate_{self.BLOCK_EARTH_MOVER_SIMILARITY}", **hash_dict}
-        cache_key_denominator = {"metric_name": f"intermediate_{self.BLOCK_EARTH_MOVER_SIMILARITY_NORM2}", **hash_dict}
+        cache_key_numerator = {"metric_name": f"intermediate_{self.BLOCK_EMD}", **hash_dict}
+        cache_key_denominator = {"metric_name": "intermediate_ems_extreme_denominator", **hash_dict}
 
         assert self._cache is not None
         emd_raw, _ = self._cache.get(cache_key_numerator, compute_numerator)
@@ -546,8 +549,9 @@ def compute():
 
         hash_dict = {
             "reference_image": str(AnnotatedImageMetrics.HASH_FUNC(ref_image, hash_size=self.HASH_LENGTH)),
+            "generated_image": str(AnnotatedImageMetrics.HASH_FUNC(pred_image, hash_size=self.HASH_LENGTH)),
         }
-        cache_key = {"metric_name": f"intermediate_{self.BLOCK_EARTH_MOVER_SIMILARITY}", **hash_dict}
+        cache_key = {"metric_name": f"intermediate_{self.BLOCK_EMD}", **hash_dict}
         assert self._cache is not None
         emd_raw, _ = self._cache.get(cache_key, compute)
 
@@ -564,8 +568,8 @@ def compute_block_emd_raw_wrapper(
         use_tqdm: bool = False,
     ):
         """Computes the block Earth Moving Distance (EMD). This attempts to
-        speed up EMD for images with huge areas by considering movement/transformatio
-        of blocks of pixels. The score is normalized against EMD against white images
+        speed up EMD for images with huge areas by considering
+        movement/transformation of blocks of pixels.
         """
         emd_value = compute_emd_recursive(
             pred_image,

src/helm/benchmark/run_specs/vlm_run_specs.py

@@ -117,9 +117,9 @@ def _get_image2structure_metric_specs(
         metric_names = [
             AnnotatedImageMetrics.PIXEL_SIMILARITY,
             AnnotatedImageMetrics.FID_SIMILARITY,
-            AnnotatedImageMetrics.BLOCK_EARTH_MOVER_SIMILARITY,
-            AnnotatedImageMetrics.BLOCK_EARTH_MOVER_SIMILARITY_NORM2,
-            AnnotatedImageMetrics.BLOCK_EARTH_MOVER_SIMILARITY_NORM1,
+            AnnotatedImageMetrics.BLOCK_EMD,
+            AnnotatedImageMetrics.EARTH_MOVER_SIMILARITY,
+            AnnotatedImageMetrics.EARTH_MOVER_SIMILARITY_WHITE,
         ]
     if include_edit_similarity:
         metric_names.append(AnnotatedImageMetrics.EDIT_SIMILARITY)

src/helm/benchmark/static/schema_image2structure.yaml

@@ -76,20 +76,20 @@ metrics:
     description: Average Levenshtein edit similarity (1 - distance normalized by length of longer sequence) between model generation and reference.
 
   # Vision Language metrics [image]:
-  - name: block_emd_similarity
-    display_name: Block Earth Mover Similarity
-    short_display_name: Block EMS
-    description: Block Earth Mover Similarity
+  - name: block_emd
+    display_name: Block Earth Mover Distance
+    short_display_name: Block-EMD
+    description: Block Earth Mover Distance (EMD adapted to speed up calculations)
+    lower_is_better: true
+  - name: earth_mover_similarity
+    display_name: Earth Mover Similarity
+    short_display_name: EMS
+    description: Earth Mover Similarity
     lower_is_better: false
-  - name: block_emd_similarity_white
-    display_name: Block Earth Mover Similarity (white)
-    short_display_name: Block EMS (white)
-    description: Block Earth Mover Similarity (white)
-    lower_is_better: false
-  - name: block_emd_similarity_median_color
-    display_name: Block Earth Mover Similarity (median)
-    short_display_name: Block EMS (median)
-    description: Block Earth Mover Similarity (median)
+  - name: earth_mover_similarity_white
+    display_name: Earth Mover Similarity (against White)
+    short_display_name: EMS (white)
+    description: Earth Mover Similarity against white image
     lower_is_better: false
   - name: pixel_similarity
     display_name: Pixel Similarity
@@ -169,11 +169,9 @@ metric_groups:
         split: ${main_split}
       - name: fid_similarity
         split: ${main_split}
-      - name: block_emd_similarity
-        split: ${main_split}
-      - name: block_emd_similarity_white
+      - name: block_emd
         split: ${main_split}
-      - name: block_emd_similarity_median_color
+      - name: earth_mover_similarity
         split: ${main_split}
 
   - name: generation_text