Changed prompt for Gemini
Browse files- mineru_single.py +12 -4
mineru_single.py
CHANGED
@@ -195,10 +195,18 @@ def call_gemini_for_image_description(image_data: bytes) -> str:
|
|
195 |
contents=[
|
196 |
{
|
197 |
"parts": [
|
198 |
-
{"text": """The provided image is a part of a question paper or markscheme.
|
199 |
-
|
200 |
-
|
201 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
202 |
{
|
203 |
"inline_data": {
|
204 |
"mime_type": "image/jpeg",
|
|
|
195 |
contents=[
|
196 |
{
|
197 |
"parts": [
|
198 |
+
{"text": """The provided image is a part of a question paper or markscheme.
|
199 |
+
Extract all the necessary information from the image to be able to identify the question.
|
200 |
+
To identify the question, we only need the following: question number and question part.
|
201 |
+
Don't include redundant information.
|
202 |
+
For example, if image contains text like: "Q1 Part A Answer: Life on earth was created by diety..."
|
203 |
+
you should return just "Q1 Part A Mark Scheme"
|
204 |
+
If there is no text on this image, return the description of the image. 20 words max.
|
205 |
+
|
206 |
+
If there are not enough data, consider information from the surrounding context.
|
207 |
+
Additionally, if the image contains a truncated part, you must describe it and mark as a
|
208 |
+
part of some another image that goes before or after current image.
|
209 |
+
"""},
|
210 |
{
|
211 |
"inline_data": {
|
212 |
"mime_type": "image/jpeg",
|