Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -115,6 +115,28 @@ will yield the image
|
|
115 |
|
116 |
The loader can then be iterated on normally and yields questions. Many questions rely on the same image, so there is some amount of data duplication.
|
117 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
118 |
### Data Splits
|
119 |
|
120 |
|
|
|
115 |
|
116 |
The loader can then be iterated on normally and yields questions. Many questions rely on the same image, so there is some amount of data duplication.
|
117 |
|
118 |
+
For this sample 0, the question has just one possible answer, but in general `answers` is a list of strings.
|
119 |
+
|
120 |
+
```python
|
121 |
+
# int identifier of the question
|
122 |
+
|
123 |
+
print(dataset["train"][0]['question_id'])
|
124 |
+
>>> 9951
|
125 |
+
|
126 |
+
# actual question
|
127 |
+
|
128 |
+
print(dataset["train"][0]['question'])
|
129 |
+
>>>'When is the contract effective date?'
|
130 |
+
|
131 |
+
# one-element list of accepted/ground truth answers for this question
|
132 |
+
|
133 |
+
print(dataset["train"][0]['answers'])
|
134 |
+
>>> ['7 - 1 - 99']
|
135 |
+
```
|
136 |
+
|
137 |
+
`ocr_results` contains OCR information about all files, which can be used for models that don't leverage only the image input.
|
138 |
+
|
139 |
+
|
140 |
### Data Splits
|
141 |
|
142 |
|