Upload folder using huggingface_hub

ZoomLDM-demo-dataset.py

@@ -3,7 +3,6 @@ import numpy as np
 from pathlib import Path
 import torch
 import torch.nn.functional as F
-from PIL import Image
 
 _DATASET_VERSION = datasets.Version("1.0.0")
 
@@ -74,7 +73,7 @@ class TCGADataset(datasets.GeneratorBasedBuilder):
             description=f"Dataset with images and SSL features. Configuration: {self.config.name}",
             features=datasets.Features(
                 {
-                    "image": datasets.Array3D(shape=(256, 256, 3), dtype="uint8"), 
+                    "image": datasets.Image(),
                     "ssl_feat": datasets.Array3D(shape=self.config.ssl_feat_shape_pooled, dtype="float32"),
                     "ssl_feat_unpooled": datasets.Array3D(shape=self.config.ssl_feat_shape_unpooled, dtype="float32"),
                     "mag": datasets.Value("int32"),
@@ -86,35 +85,30 @@ class TCGADataset(datasets.GeneratorBasedBuilder):
     def _split_generators(self, dl_manager):
 
 
-        mag_folder =  f"data/{self.config.mag_level}"  
+        original_script_dir = Path(self.base_path)
+        mag_folder_name = self.config.data_dir
+        
+        mag_data_abs_path = original_script_dir / "data" / mag_folder_name
         
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
                 gen_kwargs={
-                    "mag_folder": mag_folder, 
+                    "mag_folder_abs_path": mag_data_abs_path, 
                     "mag_level": self.config.mag_level,
                 },
             ),
         ]
 
-    def _generate_examples(self, mag_folder: Path, mag_level: str):
+    def _generate_examples(self, mag_folder_abs_path: Path, mag_level: str):
         idx = 0
         for i in range(16):
             img_filename = f"{i}.jpg"
             feat_filename = f"{i}_ssl_feat.npy"
 
-            img_path = f"{self.base_path}/{mag_folder}/{img_filename}"
-            feat_path = f"{self.base_path}/{mag_folder}/{feat_filename}"
-
-            print(f"base path: {self.base_path}")
-            print(f"mag folder: {mag_folder}")
-            print(f"img path: {img_path}")
-            print(f"feat path: {feat_path}")
-
+            img_path = mag_folder_abs_path / img_filename
+            feat_path = mag_folder_abs_path / feat_filename
 
-            img = Image.open(img_path).convert("RGB")
-            img_np = np.array(img)
 
             ssl_feat_data = np.load(feat_path)
             processed_feature = preprocess_features(ssl_feat_data)
@@ -131,7 +125,7 @@ class TCGADataset(datasets.GeneratorBasedBuilder):
 
 
             yield idx, {
-                "image": img_np, 
+                "image": str(img_path), 
                 "ssl_feat": feat_array_pooled,
                 "ssl_feat_unpooled": feat_array_unpooled,
                 "mag": _MAG_DICT[mag_level],