Update vlm-projects-multi-lang-final.py

vlm-projects-multi-lang-final.py

@@ -1,14 +1,18 @@
-
-import datasets
+# dataset.py
 import os
 import pandas as pd
+import datasets
 
 _DESCRIPTION = "A multilingual medical imaging dataset with questions and answers, structured by language."
 _HOMEPAGE = "https://huggingface.co/datasets/tungvu3196/vlm-projects-multi-lang-final"
 _LICENSE = "apache-2.0"
-_CITATION = "" # Add your citation here
+_CITATION = ""
 
-LANGUAGES = ['English', 'Vietnamese', 'French', 'German', 'Spanish', 'Russian', 'Korean', 'Mandarin', 'Japanese', 'Thai', 'Indonesian', 'Malay', 'Arabic', 'Hindi', 'Turkish', 'Portuguese']
+LANGUAGES = [
+    "English","Vietnamese","French","German","Spanish","Russian","Korean",
+    "Mandarin","Japanese","Thai","Indonesian","Malay","Arabic","Hindi",
+    "Turkish","Portuguese"
+]
 
 class VlmProjectsMultiLangFinal(datasets.GeneratorBasedBuilder):
     BUILDER_CONFIGS = [
@@ -21,33 +25,73 @@ class VlmProjectsMultiLangFinal(datasets.GeneratorBasedBuilder):
     ]
 
     def _info(self):
-        # Use the exact features we extracted from the Parquet file
         return datasets.DatasetInfo(
             description=_DESCRIPTION,
-            features={'A1': Value('string'), 'A2': Value('string'), 'A3': Value('string'), 'A4': Value('string'), 'Bbox coordinates normalized (X, Y, W, H)': Value('string'), 'Column 9': Value('float64'), 'Deliverable': Value('string'), 'Doctor': Value('string'), 'Google Drive Link': Value('string'), 'No.': Value('int64'), 'Notes': Value('string'), 'Original': Value('string'), 'Patient ID': Value('string'), 'Q1': Value('string'), 'Q2': Value('string'), 'Q3': Value('string'), 'Q4': Value('string'), 'Remove Status': Value('string'), 'Slide': Value('string'), 'Start date': Value('float64'), 'Status': Value('string'), '__index_level_0__': Value('int64'), 'image': Image(mode=None, decode=True), 'image_with_bboxes': Image(mode=None, decode=True), 'rotated_link': Value('string')},
             homepage=_HOMEPAGE,
             license=_LICENSE,
             citation=_CITATION,
+            features=datasets.Features({
+                "A1": datasets.Value("string"),
+                "A2": datasets.Value("string"),
+                "A3": datasets.Value("string"),
+                "A4": datasets.Value("string"),
+                "Bbox coordinates normalized (X, Y, W, H)": datasets.Value("string"),
+                "Column 9": datasets.Value("float64"),
+                "Deliverable": datasets.Value("string"),
+                "Doctor": datasets.Value("string"),
+                "Google Drive Link": datasets.Value("string"),
+                "No.": datasets.Value("int64"),
+                "Notes": datasets.Value("string"),
+                "Original": datasets.Value("string"),
+                "Patient ID": datasets.Value("string"),
+                "Q1": datasets.Value("string"),
+                "Q2": datasets.Value("string"),
+                "Q3": datasets.Value("string"),
+                "Q4": datasets.Value("string"),
+                "Remove Status": datasets.Value("string"),
+                "Slide": datasets.Value("string"),
+                "Start date": datasets.Value("float64"),
+                "Status": datasets.Value("string"),
+                "__index_level_0__": datasets.Value("int64"),
+                # These two will render in the Viewer if the underlying files exist in the repo:
+                "image": datasets.Image(),                 # path or dict -> file in repo
+                "image_with_bboxes": datasets.Image(),
+                # keep as string/URL if it's not a local file:
+                "rotated_link": datasets.Value("string"),
+            }),
         )
 
     def _split_generators(self, dl_manager):
-        # The data files are already in the repository, so we don't download.
-        # We just point to their relative paths.
-        lang_dir = self.config.name.lower() # 'English' -> 'english'
-        
+        # Map config name ("English") to folder ("english")
+        lang_dir = self.config.name.lower()
+        base = os.path.join(self.config.data_dir or "data", lang_dir)
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={"filepath": os.path.join("data", lang_dir, "train.parquet")},
+                gen_kwargs={"filepath": os.path.join(base, "train.parquet"),
+                            "base_dir": base},
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.TEST,
-                gen_kwargs={"filepath": os.path.join("data", lang_dir, "test.parquet")},
+                gen_kwargs={"filepath": os.path.join(base, "test.parquet"),
+                            "base_dir": base},
             ),
         ]
 
-    def _generate_examples(self, filepath):
+    def _generate_examples(self, filepath, base_dir):
+        # Read parquet produced by your pipeline
         df = pd.read_parquet(filepath)
+
         for i, row in df.iterrows():
-            yield i, row.to_dict()
+            ex = row.to_dict()
+
+            # If parquet stored relative paths like "images/xyz.png", keep them relative to repo:
+            for col in ("image", "image_with_bboxes"):
+                p = ex.get(col)
+                if isinstance(p, str) and len(p):
+                    # If the path isn't an URL, make it relative to the dataset files
+                    if not (p.startswith("http://") or p.startswith("https://")):
+                        ex[col] = os.path.join(base_dir, p).replace("\\", "/")
+                    # if it *is* a URL, leave as-is (Image will try to download)
 
+            yield i, ex