init project

app.py

@@ -503,8 +503,8 @@ def get_reconstructed_scene(outdir, filelist, schedule='linear', niter=300, min_
 
     sam2 = SAM2VideoPredictor.from_pretrained('facebook/sam2.1-hiera-large', device=device)
 
-    siglip = AutoModel.from_pretrained("google/siglip2-large-patch16-256", device_map=device)
-    siglip_processor = AutoProcessor.from_pretrained("google/siglip2-large-patch16-256")
+    siglip = AutoModel.from_pretrained("google/siglip-large-patch16-256", device_map=device).eval()
+    siglip_processor = AutoProcessor.from_pretrained("google/siglip-large-patch16-256")
 
     SAM1_DECODER_CKP = './checkpoints/Prompt_guided_Mask_Decoder.pt'
     mobilesamv2 = sam_model_registry['sam_vit_h'](None)
@@ -602,8 +602,8 @@ def get_3D_object_from_scene(outdir, text, threshold, scene, min_conf_thr=3.0, a
                  mask_sky=False, clean_depth=True, transparent_cams=True, cam_size=0.05):
     
     device = 'cpu'
-    siglip_tokenizer = AutoTokenizer.from_pretrained("google/siglip2-large-patch16-256")
-    siglip = AutoModel.from_pretrained("google/siglip2-large-patch16-256", device_map=device)
+    siglip_tokenizer = AutoTokenizer.from_pretrained("google/siglip-large-patch16-256")
+    siglip = AutoModel.from_pretrained("google/siglip-large-patch16-256", device_map=device).eval()
 
     texts = [text]
     inputs = siglip_tokenizer(text=texts, padding="max_length", return_tensors="pt")