Fix model loading: BEiT support, token fallback (HF_TOKEN), condition bug

app.py

@@ -143,21 +143,17 @@ def get_model(model_name):
                         backbone_class=tf.keras.applications.ResNet50V2,
                         nb_classes = n_classes,load_weights=False,finer_model=True,backbone_name ='Resnet50v2')
         model.load_weights('model_classification/rock-170.h5')
-    # elif model_name == 'Fossils 142': #BEiT
-    #     n_classes = 142
-    #     model = get_triplet_model_beit(input_shape = (384, 384, 3),
-    #                               embedding_units = 256,
-    #                               embedding_depth = 2,
-    #                               n_classes = n_classes)
-    #     model.load_weights('model_classification/fossil-142.h5')
-    # elif model_name == 'Fossils new': # BEiT-v2
-    #     n_classes = 142
-    #     model = get_triplet_model_beit(input_shape = (384, 384, 3),
-    #                               embedding_units = 256,
-    #                               embedding_depth = 2,
-    #                               n_classes = n_classes)
-    #     model.load_weights('model_classification/fossil-new.h5')
-    elif model_name == 'Fossils 142': # new resnet
+    elif model_name == 'Fossils BEiT':
+        n_classes = 142
+        from inference_resnet import _ensure_models_downloaded
+        from inference_beit import get_triplet_model_beit
+        _ensure_models_downloaded()
+        model = get_triplet_model_beit(input_shape=(384, 384, 3),
+                                      embedding_units=256,
+                                      embedding_depth=2,
+                                      n_classes=n_classes)
+        model.load_weights('model_classification/fossil-142.h5')
+    elif model_name == 'Fossils 142':  # resnet v2 (full model from file)
         n_classes = 142
         from inference_resnet_v2 import get_resnet_model
         model,_,_ = get_resnet_model('model_classification/fossil-model.h5')
@@ -246,7 +242,7 @@ def generate_diagram_closest(input_image,model_name,top_k):
 def explain_image(input_image,model_name,explain_method,nb_samples):
     model,n_classes= get_model(model_name)
     from explanations import explain
-    if model_name=='Fossils BEiT' or 'Fossils 142':
+    if model_name in ('Fossils BEiT', 'Fossils 142'):
         size = 384
     else:
         size = 600
@@ -559,10 +555,18 @@ custom_css = """
     margin-left: calc(50% - 50vw);
     margin-right: calc(50% - 50vw);
     padding: 48px 0 28px 0;
-    background: #14161c !important; /* force dark header in light mode */
-    color: #e9edf5 !important;
+    background: #14161c;
+    color: #e9edf5;
     border-bottom: 1px solid rgba(255,255,255,0.08);
 }
+/* Light mode overrides */
+body:not(.dark) .hero,
+[data-theme="default"] .hero,
+.gradio-container:not(.dark) .hero {
+    background: #f5f5f5;
+    color: #1a1a1a;
+    border-bottom: 1px solid rgba(0,0,0,0.1);
+}
 .hero-inner {
     max-width: 1200px;
     margin: 0 auto;
@@ -575,21 +579,41 @@ custom_css = """
     letter-spacing: -0.03em;
     color: #e9edf5;
 }
+body:not(.dark) .hero-title,
+[data-theme="default"] .hero-title,
+.gradio-container:not(.dark) .hero-title {
+    color: #1a1a1a;
+}
 .hero-subtitle {
     margin-top: 10px;
     font-size: 18px;
     color: rgba(233,237,245,0.75);
 }
+body:not(.dark) .hero-subtitle,
+[data-theme="default"] .hero-subtitle,
+.gradio-container:not(.dark) .hero-subtitle {
+    color: rgba(26,26,26,0.75);
+}
 .hero-links {
     margin-top: 18px;
     font-size: 14px;
     color: rgba(233,237,245,0.70);
 }
+body:not(.dark) .hero-links,
+[data-theme="default"] .hero-links,
+.gradio-container:not(.dark) .hero-links {
+    color: rgba(26,26,26,0.70);
+}
 .hero-links a {
     color: #9ae66e;
     text-decoration: none;
     font-weight: 600;
 }
+body:not(.dark) .hero-links a,
+[data-theme="default"] .hero-links a,
+.gradio-container:not(.dark) .hero-links a {
+    color: #2d7a32;
+}
 .hero-links a:hover {
     text-decoration: underline;
 }
@@ -597,11 +621,21 @@ custom_css = """
     margin: 0 10px;
     color: rgba(233,237,245,0.35);
 }
+body:not(.dark) .hero-sep,
+[data-theme="default"] .hero-sep,
+.gradio-container:not(.dark) .hero-sep {
+    color: rgba(26,26,26,0.35);
+}
 .hero-guide {
     margin-top: 24px;
     padding-top: 18px;
     border-top: 1px solid rgba(233,237,245,0.12);
 }
+body:not(.dark) .hero-guide,
+[data-theme="default"] .hero-guide,
+.gradio-container:not(.dark) .hero-guide {
+    border-top: 1px solid rgba(0,0,0,0.12);
+}
 .hero-guide-grid {
     display: grid;
     grid-template-columns: 1fr 1fr;
@@ -615,6 +649,11 @@ custom_css = """
     text-transform: uppercase;
     margin-bottom: 8px;
 }
+body:not(.dark) .hero-guide-h,
+[data-theme="default"] .hero-guide-h,
+.gradio-container:not(.dark) .hero-guide-h {
+    color: rgba(26,26,26,0.92);
+}
 .hero-guide-ul,
 .hero-guide-ol {
     margin: 0;
@@ -623,10 +662,26 @@ custom_css = """
     font-size: 14px;
     line-height: 1.45;
 }
+body:not(.dark) .hero-guide-ul,
+body:not(.dark) .hero-guide-ol,
+[data-theme="default"] .hero-guide-ul,
+[data-theme="default"] .hero-guide-ol,
+.gradio-container:not(.dark) .hero-guide-ul,
+.gradio-container:not(.dark) .hero-guide-ol {
+    color: rgba(26,26,26,0.78);
+}
 .hero-guide-ul b,
 .hero-guide-ol b {
     color: rgba(233,237,245,0.92);
 }
+body:not(.dark) .hero-guide-ul b,
+body:not(.dark) .hero-guide-ol b,
+[data-theme="default"] .hero-guide-ul b,
+[data-theme="default"] .hero-guide-ol b,
+.gradio-container:not(.dark) .hero-guide-ul b,
+.gradio-container:not(.dark) .hero-guide-ol b {
+    color: rgba(26,26,26,0.92);
+}
 @media (max-width: 900px) {
   .hero-title { font-size: 52px; }
   .hero-guide-grid { grid-template-columns: 1fr; }
@@ -693,8 +748,7 @@ with gr.Blocks(theme='sudeepshouche/minimalist', css=custom_css) as demo:
                 with gr.Accordion("📸 Browse Florissant fossils (non-NA)", open=True):
                     gr.Markdown(
                         "<p style='font-size: 13px; margin-bottom: 10px;'>"
-                        "These thumbnails are sourced from the Florissant unknown-fossils list "
-                        "and exclude <b>Not Applicable</b>. "
+                        "These thumbnails are sourced from the Florissant dataset, from specimens where there are doubts about their family. "
                         "For full context pages, use: "
                         "<a href='https://serre-lab.github.io/FossilLeafLens/' target='_blank'>Fossil Leaf Lens</a>."
                         "</p>"

closest_sample.py

@@ -93,13 +93,13 @@ def load_pickle_safe(filepath):
 if not os.path.exists("dataset"):
     # Avoid downloading large datasets automatically during local runs.
     # If needed, set READ_TOKEN and download manually, or run in Spaces.
-    token = os.environ.get("READ_TOKEN")
+    token = os.environ.get("READ_TOKEN") or os.environ.get("HF_TOKEN")
     if token:
         REPO_ID = "Serrelab/Fossils"
         print(f"Read token:{token}")
         snapshot_download(repo_id=REPO_ID, token=token, repo_type="dataset", local_dir="dataset")
     else:
-        print("WARNING: dataset/ not found and READ_TOKEN is not set. Closest-sample features may fail until the dataset is available.")
+        print("WARNING: dataset/ not found and READ_TOKEN (or HF_TOKEN) is not set. Closest-sample features may fail until the dataset is available.")
 
 
 fossils_pd= pd.read_csv(os.path.join(os.path.dirname(__file__), "data", "all_fossils_filtered_100.csv"))
@@ -196,7 +196,7 @@ def get_images(embedding,model_name):
         pca_leaves = load_pickle_safe('pca_leaves_170_finer.pkl')
         embedding_fossils = np.load('dataset/embedding_fossils_170_finer.npy')
         #embedding_leaves = np.load('embedding_leaves.npy')
-    elif model_name in ['Fossils 142']:
+    elif model_name in ['Fossils 142', 'Fossils BEiT']:
         pca_fossils = load_pickle_safe('pca_fossils_142_resnet.pkl')
         pca_leaves = load_pickle_safe('pca_leaves_142_resnet.pkl')
         embedding_fossils = np.load('dataset/embedding_fossils_142_finer.npy')
@@ -265,7 +265,7 @@ def get_diagram(embedding,top_k,model_name):
         pca_leaves = load_pickle_safe('pca_leaves_170_finer.pkl')
         embedding_fossils = np.load('dataset/embedding_fossils_170_finer.npy')
         #embedding_leaves = np.load('embedding_leaves.npy')
-    elif model_name in ['Fossils 142']:
+    elif model_name in ['Fossils 142', 'Fossils BEiT']:
         pca_fossils = load_pickle_safe('pca_fossils_142_resnet.pkl')
         pca_leaves = load_pickle_safe('pca_leaves_142_resnet.pkl')
         embedding_fossils = np.load('dataset/embedding_fossils_142_finer.npy')

inference_resnet.py

@@ -22,11 +22,11 @@ def _ensure_models_downloaded():
     if os.path.exists("model_classification"):
         return
     REPO_ID = "Serrelab/fossil_classification_models"
-    token = os.getenv("READ_TOKEN")
+    token = os.getenv("READ_TOKEN") or os.getenv("HF_TOKEN")
     if token is None:
         raise RuntimeError(
-            "model_classification/ is missing and READ_TOKEN is not set. "
-            "Set READ_TOKEN in .env to download models, or copy models into model_classification/."
+            "model_classification/ is missing and READ_TOKEN (or HF_TOKEN) is not set. "
+            "Set READ_TOKEN in .env or HF_TOKEN on Spaces to download models."
         )
     print("read token:", token)
     snapshot_download(repo_id=REPO_ID, token=token, repo_type="model", local_dir="model_classification")

inference_resnet_v2.py

@@ -22,11 +22,11 @@ def _ensure_models_downloaded():
     if os.path.exists("model_classification"):
         return
     REPO_ID = "Serrelab/fossil_classification_models"
-    token = os.getenv("READ_TOKEN")
+    token = os.getenv("READ_TOKEN") or os.getenv("HF_TOKEN")
     if token is None:
         raise RuntimeError(
-            "model_classification/ is missing and READ_TOKEN is not set. "
-            "Set READ_TOKEN in .env to download models, or copy models into model_classification/."
+            "model_classification/ is missing and READ_TOKEN (or HF_TOKEN) is not set. "
+            "Set READ_TOKEN in .env or HF_TOKEN on Spaces to download models."
         )
     print("read token:", token)
     snapshot_download(repo_id=REPO_ID, token=token, repo_type="model", local_dir="model_classification")

inference_sam.py

@@ -29,11 +29,11 @@ def _ensure_sam_downloaded():
   if os.path.exists("model"):
     return
   REPO_ID = "Serrelab/SAM_Leaves"
-  token = os.environ.get("READ_TOKEN")
+  token = os.environ.get("READ_TOKEN") or os.environ.get("HF_TOKEN")
   if token is None:
     raise RuntimeError(
-      "model/ is missing and READ_TOKEN is not set. "
-      "Set READ_TOKEN in .env to download SAM weights, or copy them into model/."
+      "model/ is missing and READ_TOKEN (or HF_TOKEN) is not set. "
+      "Set READ_TOKEN in .env or HF_TOKEN on Spaces to download SAM weights."
     )
   print(f"Read token:{token}")
   snapshot_download(repo_id=REPO_ID, token=token, repo_type="model", local_dir="model")