Spaces:
Running
Running
Linoy Tsaban
commited on
Commit
•
5c86655
1
Parent(s):
f58d8b8
Update app.py
Browse filesfix auto-inversion and auto-caption when removing an image
app.py
CHANGED
@@ -27,12 +27,14 @@ blip_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image
|
|
27 |
## IMAGE CPATIONING ##
|
28 |
def caption_image(input_image):
|
29 |
|
30 |
-
|
31 |
-
|
|
|
32 |
|
33 |
-
|
34 |
-
|
35 |
-
|
|
|
36 |
|
37 |
|
38 |
## DDPM INVERSION AND SAMPLING ##
|
@@ -361,8 +363,10 @@ with gr.Blocks(css="style.css") as demo:
|
|
361 |
|
362 |
|
363 |
def reset_do_inversion():
|
364 |
-
|
365 |
-
|
|
|
|
|
366 |
|
367 |
def reset_do_reconstruction():
|
368 |
do_reconstruction = True
|
@@ -597,6 +601,7 @@ with gr.Blocks(css="style.css") as demo:
|
|
597 |
# Automatically start inverting upon input_image change
|
598 |
input_image.change(
|
599 |
fn = reset_do_inversion,
|
|
|
600 |
outputs = [do_inversion],
|
601 |
queue = False).then(fn = caption_image,
|
602 |
inputs = [input_image],
|
|
|
27 |
## IMAGE CPATIONING ##
|
28 |
def caption_image(input_image):
|
29 |
|
30 |
+
if not input_image is None:
|
31 |
+
inputs = blip_processor(images=input_image, return_tensors="pt").to(device)
|
32 |
+
pixel_values = inputs.pixel_values
|
33 |
|
34 |
+
generated_ids = blip_model.generate(pixel_values=pixel_values, max_length=50)
|
35 |
+
generated_caption = blip_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
36 |
+
return generated_caption
|
37 |
+
return ""
|
38 |
|
39 |
|
40 |
## DDPM INVERSION AND SAMPLING ##
|
|
|
363 |
|
364 |
|
365 |
def reset_do_inversion():
|
366 |
+
if not input_image is None:
|
367 |
+
return True
|
368 |
+
else:
|
369 |
+
return False
|
370 |
|
371 |
def reset_do_reconstruction():
|
372 |
do_reconstruction = True
|
|
|
601 |
# Automatically start inverting upon input_image change
|
602 |
input_image.change(
|
603 |
fn = reset_do_inversion,
|
604 |
+
inputs = [input_image],
|
605 |
outputs = [do_inversion],
|
606 |
queue = False).then(fn = caption_image,
|
607 |
inputs = [input_image],
|