fix(clip): allow bad fsw strings

signwriting_evaluation/metrics/base.py

@@ -23,7 +23,7 @@ def corpus_score(self, hypotheses: list[str], references: list[list[str]]) -> fl
 
     def score_all(self, hypotheses: list[str], references: list[str]) -> list[list[float]]:
         # Default implementation: call the score function for each hypothesis-reference pair
-        return [[self.score(h, r) for r in references] for h in tqdm(hypotheses)]
+        return [[self.score(h, r) for r in references] for h in tqdm(hypotheses, disable=len(hypotheses) == 1)]
 
     def __str__(self):
         return self.name

signwriting_evaluation/metrics/clip.py

@@ -15,9 +15,19 @@
 
 
 def signwriting_to_clip_image(signwriting: CLIPInput, size=224) -> Image:
-    img = signwriting_to_image(signwriting) if isinstance(signwriting, str) else signwriting
     new_img = Image.new('RGB', (size, size), (255, 255, 255))
 
+    if isinstance(signwriting, str):
+        try:
+            img = signwriting_to_image(signwriting)
+        except ValueError as value_error:
+            # This may happen when the M box maximum values are lower
+            # than the symbols minimum values
+            print(value_error)
+            return new_img
+    else:
+        img = signwriting
+
     if img.width > size or img.height > size:
         return new_img
 
@@ -99,7 +109,7 @@ def get_clip_features(self, inputs: list[CLIPInput]):
         missing = [clip_input for clip_input in inputs if self.cache_name(clip_input) not in self.cached_texts]
 
         if len(missing) > 0:
-            pbar_disable = len(missing) < self.batch_size
+            pbar_disable = len(missing) <= self.batch_size
             pbar = tqdm(total=len(inputs), initial=len(inputs) - len(missing),
                         desc="Computing CLIP features", disable=pbar_disable)
 
@@ -112,7 +122,7 @@ def get_clip_features(self, inputs: list[CLIPInput]):
 
             pbar.close()
 
-        texts = tqdm(inputs, desc="Loading features cache", disable=len(inputs) < self.batch_size)
+        texts = tqdm(inputs, desc="Loading features cache", disable=len(inputs) <= self.batch_size)
         cached_features = [self.cache[self.cache_name(text)].cpu() for text in texts]
         features = torch.stack(cached_features)
 

signwriting_evaluation/metrics/test_clip.py

@@ -1,8 +1,9 @@
 import unittest
 
+import numpy as np
 from PIL import Image
 
-from signwriting_evaluation.metrics.clip import SignWritingCLIPScore
+from signwriting_evaluation.metrics.clip import SignWritingCLIPScore, signwriting_to_clip_image
 
 
 class TestSignWritingCLIPScore(unittest.TestCase):
@@ -23,6 +24,10 @@ def test_score_image(self):
         self.assertIsInstance(score, float)  # Check if the score is a float
         self.assertAlmostEqual(score, 0.7759, places=2)
 
+    def test_bad_fsw_is_empty_image(self):
+        fsw = "M530x538S37602531x539"
+        image = signwriting_to_clip_image(fsw)
+        self.assertTrue(np.alltrue(np.array(image) == 255))
 
 if __name__ == '__main__':
     unittest.main()