Correct bioclipv2 results

compute_accuracy.py

@@ -14,7 +14,6 @@ SPECIES_MAP = OrderedDict([
     (6, "Mountain Lion"),     # puma concolor
     (101, "Common Eland"),    # tragelaphus oryx
     (102, "Waterbuck"),       # kobus ellipsiprymnus
-    (163, "African Wild Dog") # lycaon pictus
 ])
 
 def load_ground_truth():

hf_zeroshot.py

@@ -46,6 +46,8 @@ DESCRIPTIVE_CLASS_NAMES = [
 MODELS = [
     "openai/clip-vit-large-patch14",
     "google/siglip2-large-patch16-384",
+    "google/siglip2-large-patch16-512",
+    "google/siglip2-so400m-patch16-naflex",
     "imageomics/bioclip",
     "imageomics/bioclip-2",
     "facebook/PE-Core-L14-336",
@@ -200,13 +202,11 @@ def run_openclip_inference(model_name, image_paths, class_names):
         print(f"Error loading OpenCLIP model {model_name}: {e}")
         return None
 
-def run_siglip_inference(image_paths, class_names):
+def run_siglip_inference(model_name, image_paths, class_names):
     """Run zero-shot inference using SigLIP with manual CLIP-style computation."""
-    print("Loading SigLIP model...")
+    print(f"Loading SigLIP model: {model_name}")
     try:
         from transformers import AutoProcessor, AutoModel
-
-        model_name = "google/siglip2-so400m-patch16-naflex"
         processor = AutoProcessor.from_pretrained(model_name)
         model = AutoModel.from_pretrained(model_name)
 
@@ -228,7 +228,8 @@ def run_siglip_inference(image_paths, class_names):
                         text=prompts,
                         images=image,
                         return_tensors="pt",
-                        padding="max_length"
+                        padding="max_length",
+                        truncation=True
                     ).to(device)
 
                     outputs = model(**inputs)
@@ -333,8 +334,8 @@ def main():
         # Handle different models with appropriate methods
         if model_name in ["imageomics/bioclip", "imageomics/bioclip-2"]:
             results = run_bioclip_inference(model_name, image_paths, CLASS_NAMES)
-        elif model_name == "google/siglip2-so400m-patch16-naflex":
-            results = run_siglip_inference(image_paths, CLASS_NAMES)
+        elif model_name.startswith("google/siglip"):
+            results = run_siglip_inference(model_name, image_paths, CLASS_NAMES)
         elif model_name in ["facebook/PE-Core-L14-336", "laion/CLIP-ViT-L-14-laion2B-s32B-b82K"]:
             results = run_openclip_inference(model_name, image_paths, CLASS_NAMES)
         else:

iwildcam_demo.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa01d98dedd17fda7e2451e48884409f725a4e0c10c9066139b7ebcfb4724592
-size 127187
+oid sha256:f31fb9889d96fcecb5899a12e03102cbebac1d02a1689770ca734d748333286c
+size 272

models.txt

@@ -1,5 +1,5 @@
 facebook/PE-Core-L14-336
 google/siglip2-so400m-patch16-naflex
 openai/clip-vit-large-patch14
-imageomics/bioclip-2
+imageomics/bioclip-v2
 laion/CLIP-ViT-L-14-laion2B-s32B-b82K

process_iwildcam_data.py

@@ -52,11 +52,11 @@ def create_tensors():
 
     # Load model predictions
     model_files = [
-	'zeroshot_results_facebook_PE_Core_L14_336.json',
-        'zeroshot_results_google_siglip2_large_patch16_384.json',
-	'zeroshot_results_openai_clip_vit_large_patch14.json',
-        'zeroshot_results_imageomics_bioclip_2.json',
-	'zeroshot_results_laion_CLIP_ViT_L_14_laion2B_s32B_b82K.json',
+	    'zeroshot_results_facebook_PE_Core_L14_336.json',
+        'zeroshot_results_google_siglip2_so400m_patch16_naflex.json',
+	    'zeroshot_results_openai_clip_vit_large_patch14.json',
+        'zeroshot_results_imageomics_bioclip_v2.json',
+	    'zeroshot_results_laion_CLIP_ViT_L_14_laion2B_s32B_b82K.json',
     ]
     models, all_predictions, class_names = load_model_predictions(model_files)