Spaces:
Running
Running
Rishi Desai
commited on
Commit
·
dd4f3b1
1
Parent(s):
dc6215b
breaking file up
Browse files- caption.py +127 -103
caption.py
CHANGED
@@ -62,124 +62,148 @@ class CaptioningError(Exception):
|
|
62 |
"""Exception raised for errors in the captioning process."""
|
63 |
pass
|
64 |
|
65 |
-
def
|
66 |
-
|
67 |
image_strings = []
|
68 |
for image in images:
|
69 |
buffered = io.BytesIO()
|
70 |
image.save(buffered, format="PNG")
|
71 |
img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
72 |
image_strings.append(img_str)
|
73 |
-
|
74 |
-
|
|
|
|
|
75 |
api_key = os.environ.get("TOGETHER_API_KEY")
|
76 |
if not api_key:
|
77 |
raise ValueError("TOGETHER_API_KEY is not set in the environment.")
|
|
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
|
|
|
|
|
|
|
|
|
|
|
82 |
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
# Send the batch request
|
94 |
-
messages = [
|
95 |
-
{"role": "system", "content": get_prompt()},
|
96 |
-
{"role": "user", "content": content}
|
97 |
-
]
|
98 |
-
|
99 |
-
response = client.chat.completions.create(
|
100 |
-
model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
|
101 |
-
messages=messages
|
102 |
-
)
|
103 |
-
|
104 |
-
# Parse the response to extract captions for each image
|
105 |
-
full_response = response.choices[0].message.content.strip()
|
106 |
-
lines = full_response.splitlines()
|
107 |
-
|
108 |
-
# Extract captions from the response
|
109 |
-
image_count = len(image_strings)
|
110 |
-
captions = [""] * image_count # Initialize with empty strings
|
111 |
-
|
112 |
-
# Extract lines that start with or contain "tr1gger"
|
113 |
-
tr1gger_lines = [line for line in lines if "tr1gger" in line]
|
114 |
-
|
115 |
-
# Assign captions to images
|
116 |
-
for i in range(image_count):
|
117 |
-
if i < len(tr1gger_lines):
|
118 |
-
caption = tr1gger_lines[i]
|
119 |
-
# If caption contains but doesn't start with tr1gger, extract just that part
|
120 |
-
if not caption.startswith("tr1gger") and "tr1gger" in caption:
|
121 |
-
caption = caption[caption.index("tr1gger"):]
|
122 |
-
captions[i] = caption
|
123 |
-
|
124 |
-
# Check if all captions are empty or don't contain the trigger word
|
125 |
-
valid_captions = [c for c in captions if c and "tr1gger" in c]
|
126 |
-
if not valid_captions:
|
127 |
-
error_msg = "Failed to parse any valid captions from batch response. Response contained no lines with 'tr1gger'"
|
128 |
-
error_msg += f"\n\nActual response:\n{full_response}"
|
129 |
-
raise CaptioningError(error_msg)
|
130 |
-
|
131 |
-
# Check if some captions are missing
|
132 |
-
if len(valid_captions) < len(images):
|
133 |
-
missing_count = len(images) - len(valid_captions)
|
134 |
-
invalid_captions = [(i, c) for i, c in enumerate(captions) if not c or "tr1gger" not in c]
|
135 |
-
error_msg = f"Failed to parse captions for {missing_count} of {len(images)} images in batch mode"
|
136 |
-
error_msg += "\n\nMalformed captions:"
|
137 |
-
for idx, caption in invalid_captions:
|
138 |
-
error_msg += f"\nImage {idx+1}: '{caption}'"
|
139 |
-
raise CaptioningError(error_msg)
|
140 |
-
else:
|
141 |
-
# Original method: process each image separately
|
142 |
-
for img_str in image_strings:
|
143 |
-
messages = [
|
144 |
-
{"role": "system", "content": get_prompt()},
|
145 |
-
{
|
146 |
-
"role": "user",
|
147 |
-
"content": [
|
148 |
-
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{img_str}"}},
|
149 |
-
{"type": "text", "text": "Describe this image."}
|
150 |
-
]
|
151 |
-
}
|
152 |
]
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
180 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
181 |
return captions
|
182 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
def extract_captions(file_path):
|
184 |
captions = []
|
185 |
with open(file_path, 'r') as file:
|
|
|
62 |
"""Exception raised for errors in the captioning process."""
|
63 |
pass
|
64 |
|
65 |
+
def images_to_base64(images):
|
66 |
+
"""Convert a list of PIL images to base64 encoded strings."""
|
67 |
image_strings = []
|
68 |
for image in images:
|
69 |
buffered = io.BytesIO()
|
70 |
image.save(buffered, format="PNG")
|
71 |
img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
72 |
image_strings.append(img_str)
|
73 |
+
return image_strings
|
74 |
+
|
75 |
+
def get_together_client():
|
76 |
+
"""Initialize and return the Together API client."""
|
77 |
api_key = os.environ.get("TOGETHER_API_KEY")
|
78 |
if not api_key:
|
79 |
raise ValueError("TOGETHER_API_KEY is not set in the environment.")
|
80 |
+
return Together(api_key=api_key)
|
81 |
|
82 |
+
def extract_trigger_caption(line):
|
83 |
+
"""Extract 'tr1gger' caption from a line of text."""
|
84 |
+
if "tr1gger" in line:
|
85 |
+
# If caption doesn't start with tr1gger but contains it, extract just that part
|
86 |
+
if not line.startswith("tr1gger"):
|
87 |
+
return line[line.index("tr1gger"):]
|
88 |
+
return line
|
89 |
+
return ""
|
90 |
|
91 |
+
def caption_single_image(client, img_str):
|
92 |
+
"""Process and caption a single image."""
|
93 |
+
messages = [
|
94 |
+
{"role": "system", "content": get_prompt()},
|
95 |
+
{
|
96 |
+
"role": "user",
|
97 |
+
"content": [
|
98 |
+
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{img_str}"}},
|
99 |
+
{"type": "text", "text": "Caption this image."}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
]
|
101 |
+
}
|
102 |
+
]
|
103 |
+
|
104 |
+
# Request caption for the image using Llama 4 Maverick
|
105 |
+
response = client.chat.completions.create(
|
106 |
+
model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
|
107 |
+
messages=messages
|
108 |
+
)
|
109 |
+
|
110 |
+
# Extract caption from the response
|
111 |
+
full_response = response.choices[0].message.content.strip()
|
112 |
+
|
113 |
+
# Look for the trigger line in the response
|
114 |
+
caption = ""
|
115 |
+
for line in full_response.splitlines():
|
116 |
+
caption = extract_trigger_caption(line)
|
117 |
+
if caption:
|
118 |
+
break
|
119 |
+
|
120 |
+
# Check if caption is valid
|
121 |
+
if not caption:
|
122 |
+
error_msg = "Failed to extract a valid caption (containing 'tr1gger') from the response"
|
123 |
+
error_msg += f"\n\nActual response:\n{full_response}"
|
124 |
+
raise CaptioningError(error_msg)
|
125 |
+
|
126 |
+
return caption
|
127 |
+
|
128 |
+
def caption_batch_images(client, image_strings, category):
|
129 |
+
"""Process and caption multiple images in a single batch request."""
|
130 |
+
# Create a content array with all images
|
131 |
+
content = [{"type": "text", "text": f"Here is the batch of images for {category}. Please caption each image on a separate line, starting each caption with 'tr1gger'."}]
|
132 |
+
|
133 |
+
# Add all images to the content array
|
134 |
+
for i, img_str in enumerate(image_strings):
|
135 |
+
content.append({"type": "image_url", "image_url": {"url": f"data:image/png;base64,{img_str}"}})
|
136 |
+
content.append({"type": "text", "text": f"Image {i+1}"})
|
137 |
+
|
138 |
+
# Send the batch request
|
139 |
+
messages = [
|
140 |
+
{"role": "system", "content": get_prompt()},
|
141 |
+
{"role": "user", "content": content}
|
142 |
+
]
|
143 |
|
144 |
+
response = client.chat.completions.create(
|
145 |
+
model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
|
146 |
+
messages=messages
|
147 |
+
)
|
148 |
+
|
149 |
+
return process_batch_response(response, image_strings)
|
150 |
+
|
151 |
+
def process_batch_response(response, image_strings):
|
152 |
+
"""Process the API response from a batch request and extract captions."""
|
153 |
+
# Parse the response to extract captions for each image
|
154 |
+
full_response = response.choices[0].message.content.strip()
|
155 |
+
lines = full_response.splitlines()
|
156 |
+
|
157 |
+
# Extract captions from the response
|
158 |
+
image_count = len(image_strings)
|
159 |
+
captions = [""] * image_count # Initialize with empty strings
|
160 |
+
|
161 |
+
# Extract lines that start with or contain "tr1gger"
|
162 |
+
tr1gger_lines = [line for line in lines if "tr1gger" in line]
|
163 |
+
|
164 |
+
# Assign captions to images
|
165 |
+
for i in range(image_count):
|
166 |
+
if i < len(tr1gger_lines):
|
167 |
+
caption = extract_trigger_caption(tr1gger_lines[i])
|
168 |
+
captions[i] = caption
|
169 |
+
|
170 |
+
validate_batch_captions(captions, image_count, full_response)
|
171 |
return captions
|
172 |
|
173 |
+
def validate_batch_captions(captions, image_count, full_response):
|
174 |
+
"""Validate captions extracted from a batch response."""
|
175 |
+
# Check if all captions are empty or don't contain the trigger word
|
176 |
+
valid_captions = [c for c in captions if c and "tr1gger" in c]
|
177 |
+
if not valid_captions:
|
178 |
+
error_msg = "Failed to parse any valid captions from batch response. Response contained no lines with 'tr1gger'"
|
179 |
+
error_msg += f"\n\nActual response:\n{full_response}"
|
180 |
+
raise CaptioningError(error_msg)
|
181 |
+
|
182 |
+
# Check if some captions are missing
|
183 |
+
if len(valid_captions) < image_count:
|
184 |
+
missing_count = image_count - len(valid_captions)
|
185 |
+
invalid_captions = [(i, c) for i, c in enumerate(captions) if not c or "tr1gger" not in c]
|
186 |
+
error_msg = f"Failed to parse captions for {missing_count} of {image_count} images in batch mode"
|
187 |
+
error_msg += "\n\nMalformed captions:"
|
188 |
+
for idx, caption in invalid_captions:
|
189 |
+
error_msg += f"\nImage {idx+1}: '{caption}'"
|
190 |
+
raise CaptioningError(error_msg)
|
191 |
+
|
192 |
+
def caption_images(images, category=None, batch_mode=False):
|
193 |
+
"""Caption a list of images, either individually or in batch mode."""
|
194 |
+
# Convert PIL images to base64 encoded strings
|
195 |
+
image_strings = images_to_base64(images)
|
196 |
+
|
197 |
+
# Initialize the API client
|
198 |
+
client = get_together_client()
|
199 |
+
|
200 |
+
# Process images based on the mode
|
201 |
+
if batch_mode and category:
|
202 |
+
return caption_batch_images(client, image_strings, category)
|
203 |
+
else:
|
204 |
+
# Process each image individually
|
205 |
+
return [caption_single_image(client, img_str) for img_str in image_strings]
|
206 |
+
|
207 |
def extract_captions(file_path):
|
208 |
captions = []
|
209 |
with open(file_path, 'r') as file:
|