fixing double image triggers

app.py

@@ -51,17 +51,22 @@ from transformers import pipeline
 captioner = pipeline("image-to-text",model="Salesforce/blip-image-captioning-base")
 
 import PIL
-import numpy
 # an image has been selected. it comes to this fn as a numpy ndarray
 # convert it to a PIL image and feed to the captioner
 # return the resulting caption
-def image_supplied(img: numpy.ndarray):
-    if img is None: return
+#
+# NOTE: due to an error in gradio, this fn is triggered twice for each img change
+# only process the first fn call, and keep the first caption result
+async def image_supplied(img, count: int, last_cap: str):
+    if img is None: return "", count, ""
+    count += 1
+    if (count & 1) == 0: return last_cap, count, last_cap
     if img.any():
         im = PIL.Image.fromarray(img)
         caption = captioner(im, max_new_tokens=20)
         result = caption[0]['generated_text']
-        return result
+        #print(f"caption={result}")
+        return result, count, result
 
 # class wrapping the chat
 class ChatWrapper:
@@ -184,11 +189,14 @@ with block:
     #submit.click(chat, inputs=[openai_api_key_textbox, result_box, state, agent_state], outputs=[caption_box, state])
     #result_box.submit(chat, inputs=[openai_api_key_textbox, result_box, state, agent_state], outputs=[caption_box, state])
 
+    count = gr.State(value=0)
+    last_cap = gr.State(value="")
+
     # if image has changed, feed it to "image_supplied", and pass result to "result_box"
     image_box.change(
         image_supplied,
-        inputs=[image_box],
-        outputs=[result_box]
+        inputs=[image_box, count, last_cap],
+        outputs=[result_box, count, last_cap]
     )
 
     # if api key in input box has changed, update the key in app

requirements.txt

@@ -1,5 +1,5 @@
 openai
-gradio
+gradio>=3.35.2
 langchain
 transformers
 torch