Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -19,48 +19,42 @@ load_dotenv()
|
|
19 |
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
|
20 |
|
21 |
|
22 |
-
def
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
if not isinstance(image, Image.Image):
|
25 |
-
raise ValueError("Input must be a PIL Image")
|
26 |
-
|
27 |
-
width, height = image.size
|
28 |
-
if width > max_width or height > max_height:
|
29 |
-
aspect_ratio = width / height
|
30 |
-
if aspect_ratio > 1:
|
31 |
-
new_width = max_width
|
32 |
-
new_height = int(new_width / aspect_ratio)
|
33 |
-
else:
|
34 |
-
new_height = max_height
|
35 |
-
new_width = int(new_height * aspect_ratio)
|
36 |
-
image = image.resize((new_width, new_height), Image.Resampling.LANCZOS)
|
37 |
|
38 |
buffered = io.BytesIO()
|
39 |
-
|
40 |
-
image.save(buffered, format="JPEG", quality=quality)
|
41 |
-
buffered.seek(0)
|
42 |
return base64.b64encode(buffered.getvalue()).decode('utf-8')
|
43 |
|
|
|
44 |
def analyze_image(image):
|
45 |
client = OpenAI(api_key=OPENAI_API_KEY)
|
|
|
46 |
|
47 |
-
#
|
48 |
-
|
49 |
-
|
50 |
-
# Build the list-of-dicts prompt
|
51 |
-
prompt_dict = [
|
52 |
{
|
53 |
"type": "text",
|
54 |
-
"text": """Your task is to determine if the image is surprising or not.
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
{
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
}
|
63 |
-
"""
|
64 |
},
|
65 |
{
|
66 |
"type": "image_url",
|
@@ -70,27 +64,29 @@ def analyze_image(image):
|
|
70 |
}
|
71 |
]
|
72 |
|
73 |
-
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
-
# Send request
|
77 |
response = client.chat.completions.create(
|
78 |
-
model="gpt-4o-mini",
|
79 |
-
messages=
|
80 |
-
{
|
81 |
-
"role": "user",
|
82 |
-
"content": json_prompt
|
83 |
-
}
|
84 |
-
],
|
85 |
max_tokens=100,
|
86 |
temperature=0.1,
|
87 |
-
response_format={
|
|
|
|
|
88 |
)
|
89 |
|
90 |
return response.choices[0].message.content
|
91 |
|
92 |
|
93 |
-
|
94 |
def show_mask(mask, ax, random_color=False):
|
95 |
if random_color:
|
96 |
color = np.concatenate([np.random.random(3), np.array([0.6])], axis=0)
|
@@ -114,7 +110,7 @@ def process_image_detection(image, target_label, surprise_rating):
|
|
114 |
original_size = image.size
|
115 |
|
116 |
# Calculate relative font size based on image dimensions
|
117 |
-
base_fontsize = min(original_size) / 40 # Adjust this divisor
|
118 |
|
119 |
owlv2_processor = Owlv2Processor.from_pretrained("google/owlv2-base-patch16")
|
120 |
owlv2_model = Owlv2ForObjectDetection.from_pretrained("google/owlv2-base-patch16").to(device)
|
@@ -137,6 +133,7 @@ def process_image_detection(image, target_label, surprise_rating):
|
|
137 |
|
138 |
ax = plt.Axes(fig, [0., 0., 1., 1.])
|
139 |
fig.add_axes(ax)
|
|
|
140 |
plt.imshow(image)
|
141 |
|
142 |
scores = results["scores"]
|
@@ -165,7 +162,7 @@ def process_image_detection(image, target_label, surprise_rating):
|
|
165 |
mask = masks[0].numpy() if isinstance(masks[0], torch.Tensor) else masks[0]
|
166 |
show_mask(mask, ax=ax)
|
167 |
|
168 |
-
# Draw rectangle
|
169 |
rect = patches.Rectangle(
|
170 |
(box[0], box[1]),
|
171 |
box[2] - box[0],
|
@@ -176,7 +173,7 @@ def process_image_detection(image, target_label, surprise_rating):
|
|
176 |
)
|
177 |
ax.add_patch(rect)
|
178 |
|
179 |
-
#
|
180 |
plt.text(
|
181 |
box[0], box[1] - base_fontsize,
|
182 |
f'{max_score:.2f}',
|
@@ -186,7 +183,7 @@ def process_image_detection(image, target_label, surprise_rating):
|
|
186 |
bbox=dict(facecolor='white', alpha=0.7, edgecolor='none', pad=2)
|
187 |
)
|
188 |
|
189 |
-
#
|
190 |
plt.text(
|
191 |
box[2] + base_fontsize / 2, box[1],
|
192 |
f'Unexpected (Rating: {surprise_rating}/5)\n{target_label}',
|
@@ -199,20 +196,17 @@ def process_image_detection(image, target_label, surprise_rating):
|
|
199 |
|
200 |
plt.axis('off')
|
201 |
|
202 |
-
# Save figure to buffer
|
203 |
buf = io.BytesIO()
|
204 |
-
plt.savefig(
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
metadata={'dpi': original_dpi}
|
211 |
-
)
|
212 |
buf.seek(0)
|
213 |
plt.close()
|
214 |
|
215 |
-
#
|
216 |
output_image = Image.open(buf)
|
217 |
output_image = output_image.resize(original_size, Image.Resampling.LANCZOS)
|
218 |
|
@@ -233,17 +227,16 @@ def process_and_analyze(image):
|
|
233 |
try:
|
234 |
# Handle different input types
|
235 |
if isinstance(image, tuple):
|
236 |
-
image = image[0]
|
237 |
if isinstance(image, np.ndarray):
|
238 |
image = Image.fromarray(image)
|
239 |
if not isinstance(image, Image.Image):
|
240 |
raise ValueError("Invalid image format")
|
241 |
|
242 |
-
# Analyze image
|
243 |
gpt_response = analyze_image(image)
|
244 |
response_data = json.loads(gpt_response)
|
245 |
|
246 |
-
# If surprising, try to detect the element
|
247 |
if response_data["label"].lower() == "surprising" and response_data["element"].lower() != "na":
|
248 |
result_buf = process_image_detection(image, response_data["element"], response_data["rating"])
|
249 |
result_image = Image.open(result_buf)
|
@@ -254,7 +247,6 @@ def process_and_analyze(image):
|
|
254 |
)
|
255 |
return result_image, analysis_text
|
256 |
else:
|
257 |
-
# If not surprising or element=NA
|
258 |
return image, "Not Surprising"
|
259 |
|
260 |
except Exception as e:
|
|
|
19 |
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
|
20 |
|
21 |
|
22 |
+
def encode_image_to_base64(image):
|
23 |
+
# If image is a tuple (as sometimes provided by Gradio), take the first element
|
24 |
+
if isinstance(image, tuple):
|
25 |
+
image = image[0]
|
26 |
+
|
27 |
+
# If image is a numpy array, convert to PIL Image
|
28 |
+
if isinstance(image, np.ndarray):
|
29 |
+
image = Image.fromarray(image)
|
30 |
+
|
31 |
+
# Ensure image is in PIL Image format
|
32 |
if not isinstance(image, Image.Image):
|
33 |
+
raise ValueError("Input must be a PIL Image, numpy array, or tuple containing an image")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
buffered = io.BytesIO()
|
36 |
+
image.save(buffered, format="PNG")
|
|
|
|
|
37 |
return base64.b64encode(buffered.getvalue()).decode('utf-8')
|
38 |
|
39 |
+
|
40 |
def analyze_image(image):
|
41 |
client = OpenAI(api_key=OPENAI_API_KEY)
|
42 |
+
base64_image = encode_image_to_base64(image)
|
43 |
|
44 |
+
# --- MINIMAL FIX START ---
|
45 |
+
# We build a Python list of dicts, then JSON-encode it:
|
46 |
+
prompt_list = [
|
|
|
|
|
47 |
{
|
48 |
"type": "text",
|
49 |
+
"text": """Your task is to determine if the image is surprising or not surprising.
|
50 |
+
if the image is surprising, determine which element, figure or object in the image is making the image surprising and write it only in one sentence with no more then 6 words, otherwise, write 'NA'.
|
51 |
+
Also rate how surprising the image is on a scale of 1-5, where 1 is not surprising at all and 5 is highly surprising.
|
52 |
+
Provide the response as a JSON with the following structure:
|
53 |
{
|
54 |
+
"label": "[surprising OR not surprising]",
|
55 |
+
"element": "[element]",
|
56 |
+
"rating": [1-5]
|
57 |
+
}"""
|
|
|
58 |
},
|
59 |
{
|
60 |
"type": "image_url",
|
|
|
64 |
}
|
65 |
]
|
66 |
|
67 |
+
prompt_json = json.dumps(prompt_list)
|
68 |
+
|
69 |
+
messages = [
|
70 |
+
{
|
71 |
+
"role": "user",
|
72 |
+
"content": prompt_json # content must be a single string
|
73 |
+
}
|
74 |
+
]
|
75 |
+
# --- MINIMAL FIX END ---
|
76 |
|
|
|
77 |
response = client.chat.completions.create(
|
78 |
+
model="gpt-4o-mini", # or whichever model you have access to
|
79 |
+
messages=messages,
|
|
|
|
|
|
|
|
|
|
|
80 |
max_tokens=100,
|
81 |
temperature=0.1,
|
82 |
+
response_format={
|
83 |
+
"type": "json_object"
|
84 |
+
}
|
85 |
)
|
86 |
|
87 |
return response.choices[0].message.content
|
88 |
|
89 |
|
|
|
90 |
def show_mask(mask, ax, random_color=False):
|
91 |
if random_color:
|
92 |
color = np.concatenate([np.random.random(3), np.array([0.6])], axis=0)
|
|
|
110 |
original_size = image.size
|
111 |
|
112 |
# Calculate relative font size based on image dimensions
|
113 |
+
base_fontsize = min(original_size) / 40 # Adjust this divisor to change overall font size
|
114 |
|
115 |
owlv2_processor = Owlv2Processor.from_pretrained("google/owlv2-base-patch16")
|
116 |
owlv2_model = Owlv2ForObjectDetection.from_pretrained("google/owlv2-base-patch16").to(device)
|
|
|
133 |
|
134 |
ax = plt.Axes(fig, [0., 0., 1., 1.])
|
135 |
fig.add_axes(ax)
|
136 |
+
|
137 |
plt.imshow(image)
|
138 |
|
139 |
scores = results["scores"]
|
|
|
162 |
mask = masks[0].numpy() if isinstance(masks[0], torch.Tensor) else masks[0]
|
163 |
show_mask(mask, ax=ax)
|
164 |
|
165 |
+
# Draw rectangle with increased line width
|
166 |
rect = patches.Rectangle(
|
167 |
(box[0], box[1]),
|
168 |
box[2] - box[0],
|
|
|
173 |
)
|
174 |
ax.add_patch(rect)
|
175 |
|
176 |
+
# Add confidence score with improved visibility
|
177 |
plt.text(
|
178 |
box[0], box[1] - base_fontsize,
|
179 |
f'{max_score:.2f}',
|
|
|
183 |
bbox=dict(facecolor='white', alpha=0.7, edgecolor='none', pad=2)
|
184 |
)
|
185 |
|
186 |
+
# Add label and rating with improved visibility
|
187 |
plt.text(
|
188 |
box[2] + base_fontsize / 2, box[1],
|
189 |
f'Unexpected (Rating: {surprise_rating}/5)\n{target_label}',
|
|
|
196 |
|
197 |
plt.axis('off')
|
198 |
|
|
|
199 |
buf = io.BytesIO()
|
200 |
+
plt.savefig(buf,
|
201 |
+
format='png',
|
202 |
+
dpi=dpi,
|
203 |
+
bbox_inches='tight',
|
204 |
+
pad_inches=0,
|
205 |
+
metadata={'dpi': original_dpi})
|
|
|
|
|
206 |
buf.seek(0)
|
207 |
plt.close()
|
208 |
|
209 |
+
# Process final image
|
210 |
output_image = Image.open(buf)
|
211 |
output_image = output_image.resize(original_size, Image.Resampling.LANCZOS)
|
212 |
|
|
|
227 |
try:
|
228 |
# Handle different input types
|
229 |
if isinstance(image, tuple):
|
230 |
+
image = image[0] # Take the first element if it's a tuple
|
231 |
if isinstance(image, np.ndarray):
|
232 |
image = Image.fromarray(image)
|
233 |
if not isinstance(image, Image.Image):
|
234 |
raise ValueError("Invalid image format")
|
235 |
|
236 |
+
# Analyze image
|
237 |
gpt_response = analyze_image(image)
|
238 |
response_data = json.loads(gpt_response)
|
239 |
|
|
|
240 |
if response_data["label"].lower() == "surprising" and response_data["element"].lower() != "na":
|
241 |
result_buf = process_image_detection(image, response_data["element"], response_data["rating"])
|
242 |
result_image = Image.open(result_buf)
|
|
|
247 |
)
|
248 |
return result_image, analysis_text
|
249 |
else:
|
|
|
250 |
return image, "Not Surprising"
|
251 |
|
252 |
except Exception as e:
|