update

app.py

@@ -19,8 +19,8 @@ import torchvision
 from huggingface_hub import HfApi, login, snapshot_download
 from PIL import Image
 
-session_token = os.environ.get("SessionToken")
-login(token=session_token)
+# session_token = os.environ.get("SessionToken")
+# login(token=session_token)
 
 csv.field_size_limit(sys.maxsize)
 
@@ -35,7 +35,7 @@ with open("imagenet-labels.json") as f:
 with open("id_to_label.json", "r") as f:
     id_to_labels = json.load(f)
 
-imagenet_training_samples_path = "imagenet_traning_samples"
+imagenet_training_samples_path = "imagenet_samples"
 
 bad_items = open("./ex2.txt", "r").read().split("\n")
 bad_items = [x.split(".")[0] for x in bad_items]
@@ -48,12 +48,12 @@ gdown.cached_download(
     url="https://huggingface.co/datasets/taesiri/imagenet_hard_review_samples/resolve/main/data.zip",
     path="./data.zip",
     quiet=False,
-    md5="8666a9b361f6eea79878be6c09701def",
+    md5="ece2720fed664e71799f316a881d4324",
 )
 
 # EXTRACT if needed
 
-if not os.path.exists("./imagenet_traning_samples") or not os.path.exists(
+if not os.path.exists("./imagenet_samples") or not os.path.exists(
     "./knn_cache_for_imagenet_hard"
 ):
     torchvision.datasets.utils.extract_archive(
@@ -106,18 +106,11 @@ def generate_dataset(username):
     if NUMBER_OF_IMAGES == 0:
         return []
 
-    # random_indices = remaining
-    # random_images = [imagenet_hard[int(i)]["image"] for i in random_indices]
-    # random_gt_ids = [imagenet_hard[int(i)]["label"] for i in random_indices]
-    # random_gt_labels = [imagenet_hard[int(x)]["english_label"] for x in random_indices]
-
     data = []
     for i, image in enumerate(remaining):
         data.append(
             {
                 "id": remaining[i],
-                # "correct_label": random_gt_labels[i],
-                # "original_id": int(random_indices[i]),
             }
         )
     return data
@@ -141,7 +134,7 @@ def string_to_image(text):
     return fig
 
 
-all_samples = glob("./imagenet_traning_samples/*.JPEG")
+all_samples = glob("./imagenet_samples/*.JPEG")
 qid_to_sample = {
     int(x.split("/")[-1].split(".")[0].split("_")[0]): x for x in all_samples
 }
@@ -367,13 +360,15 @@ with gr.Blocks(css=newcss, theme=gr.themes.Soft()) as demo:
             reject_btn = gr.Button(value="Reject")
         with gr.Row():
             query_image = gr.Image(type="pil", label="Query", elem_id="query_image")
-            with gr.Column():
-                label_plot = gr.Plot(
-                    label="Is this a correct label for this image?", type="fig"
-                )
-                training_samples = gr.Gallery(
-                    type="pil", label="Training samples", elem_id="sample_gallery"
-                )
+            with gr.Row():
+                with gr.Column(scale=1):
+                    label_plot = gr.Plot(
+                        label="Is this a correct label for this image?", type="fig"
+                    )
+                with gr.Column(scale=3):
+                    training_samples = gr.Gallery(
+                        type="pil", label="Training samples", elem_id="sample_gallery"
+                    )
 
     accept_btn.click(
         update_app,