fix threshold

app.py

@@ -17,15 +17,15 @@ hence you can get better predictions by querying the image with text templates u
 """
 demo = gr.Interface(
     query_image, 
-    inputs=[gr.Image(), "text", gr.Slider(0, 1, value=0.1)], 
+    inputs=[gr.Image(), "text"], 
     outputs="image",
     title="Zero-Shot Object Detection with OWL-ViT",
     description=description,
     examples=[
-        ["./demo_images/cats.png", "cats,ears", 0.11], 
-        ["./demo_images/demo1.jpg", "bear,soil,sea", 0.1],
-        ["./demo_images/demo2.jpg", "dog,ear,leg,eyes,tail", 0.1],
-        ["./demo_images/tanager.jpg", "wing,eyes,back,legs,tail", 0.01]
+        ["./demo_images/cats.png", "cats,ears"], 
+        ["./demo_images/demo1.jpg", "bear,soil,sea"],
+        ["./demo_images/demo2.jpg", "dog,ear,leg,eyes,tail"],
+        ["./demo_images/tanager.jpg", "wing,eyes,back,legs,tail"]
     ],
 )
 

demo.py

@@ -81,7 +81,7 @@ owlvit_processor = OwlViTProcessor.from_pretrained("google/owlvit-base-patch32")
 # run segment anything (SAM)
 sam_predictor = SamPredictor(build_sam(checkpoint="./sam_vit_h_4b8939.pth"))
 
-def query_image(img, text_prompt, box_threshold):
+def query_image(img, text_prompt):
     # load image
     if not isinstance(img, PIL.Image.Image):
         pil_img = Image.fromarray(np.uint8(img)).convert('RGB')
@@ -89,7 +89,7 @@ def query_image(img, text_prompt, box_threshold):
     text_prompt = text_prompt
     texts = text_prompt.split(",")
 
-    box_threshold = box_threshold
+    box_threshold = 0.0
 
     # run object detection model
     with torch.no_grad():