Spaces:
Runtime error
Runtime error
[yesha vyas]
commited on
Commit
·
f1f8d48
1
Parent(s):
444fa81
App modified
Browse files
app.py
CHANGED
@@ -1,27 +1,98 @@
|
|
1 |
-
|
|
|
2 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
#
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
)
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import torch
|
3 |
import gradio as gr
|
4 |
+
from threading import Thread
|
5 |
+
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
|
6 |
+
from PIL import ImageDraw
|
7 |
+
import re
|
8 |
+
from torchvision.transforms.v2 import Resize
|
9 |
|
10 |
+
parser = argparse.ArgumentParser()
|
11 |
+
parser.add_argument("--cpu", action="store_true")
|
12 |
+
args = parser.parse_args()
|
13 |
+
|
14 |
+
DEVICE = "cuda"
|
15 |
+
DTYPE = torch.float32 if DEVICE == "cpu" else torch.float16 # CPU doesn't support float16
|
16 |
+
LATEST_REVISION = "2024-05-20"
|
17 |
+
|
18 |
+
model_id = "vikhyatk/moondream2"
|
19 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=LATEST_REVISION)
|
20 |
+
moondream = AutoModelForCausalLM.from_pretrained(
|
21 |
+
model_id, trust_remote_code=True, revision=LATEST_REVISION, torch_dtype=DTYPE
|
22 |
+
).to(device=DEVICE)
|
23 |
+
|
24 |
+
moondream.eval()
|
25 |
+
|
26 |
+
|
27 |
+
def answer_question(img, prompt):
|
28 |
+
image_embeds = moondream.encode_image(img)
|
29 |
+
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True)
|
30 |
+
thread = Thread(
|
31 |
+
target=moondream.answer_question,
|
32 |
+
kwargs={
|
33 |
+
"image_embeds": image_embeds,
|
34 |
+
"question": prompt,
|
35 |
+
"tokenizer": tokenizer,
|
36 |
+
"streamer": streamer,
|
37 |
+
},
|
38 |
+
)
|
39 |
+
thread.start()
|
40 |
+
|
41 |
+
buffer = ""
|
42 |
+
for new_text in streamer:
|
43 |
+
buffer += new_text
|
44 |
+
yield buffer
|
45 |
+
|
46 |
+
|
47 |
+
def extract_floats(text):
|
48 |
+
# Regular expression to match an array of four floating point numbers
|
49 |
+
pattern = r"\[\s*(-?\d+\.\d+)\s*,\s*(-?\d+\.\d+)\s*,\s*(-?\d+\.\d+)\s*,\s*(-?\d+\.\d+)\s*\]"
|
50 |
+
match = re.search(pattern, text)
|
51 |
+
if match:
|
52 |
+
# Extract the numbers and convert them to floats
|
53 |
+
return [float(num) for num in match.groups()]
|
54 |
+
return None # Return None if no match is found
|
55 |
+
|
56 |
+
|
57 |
+
def extract_bbox(text):
|
58 |
+
bbox = None
|
59 |
+
if extract_floats(text) is not None:
|
60 |
+
x1, y1, x2, y2 = extract_floats(text)
|
61 |
+
bbox = (x1, y1, x2, y2)
|
62 |
+
return bbox
|
63 |
+
|
64 |
+
|
65 |
+
def process_answer(img, answer):
|
66 |
+
if extract_bbox(answer) is not None:
|
67 |
+
x1, y1, x2, y2 = extract_bbox(answer)
|
68 |
+
draw_image = Resize(768)(img)
|
69 |
+
width, height = draw_image.size
|
70 |
+
x1, x2 = int(x1 * width), int(x2 * width)
|
71 |
+
y1, y2 = int(y1 * height), int(y2 * height)
|
72 |
+
bbox = (x1, y1, x2, y2)
|
73 |
+
ImageDraw.Draw(draw_image).rectangle(bbox, outline="red", width=3)
|
74 |
+
return gr.update(visible=True, value=draw_image)
|
75 |
+
|
76 |
+
return gr.update(visible=False, value=None)
|
77 |
+
|
78 |
+
|
79 |
+
with gr.Blocks() as demo:
|
80 |
+
gr.Markdown(
|
81 |
+
"""
|
82 |
+
# 🌔 moondream
|
83 |
+
"""
|
84 |
+
)
|
85 |
+
with gr.Row():
|
86 |
+
prompt = gr.Textbox(label="Input Prompt", placeholder="Type here...", scale=4)
|
87 |
+
submit = gr.Button("Submit")
|
88 |
+
with gr.Row():
|
89 |
+
img = gr.Image(type="pil", label="Upload an Image")
|
90 |
+
with gr.Column():
|
91 |
+
output = gr.Markdown(label="Response")
|
92 |
+
ann = gr.Image(visible=False, label="Annotated Image")
|
93 |
+
|
94 |
+
submit.click(answer_question, [img, prompt], output)
|
95 |
+
prompt.submit(answer_question, [img, prompt], output)
|
96 |
+
output.change(process_answer, [img, output], ann, show_progress=False)
|
97 |
+
|
98 |
+
demo.queue().launch(debug=True)
|