Update README.md
Browse files
README.md
CHANGED
|
@@ -7,6 +7,8 @@ task_categories:
|
|
| 7 |
pretty_name: VSR (Parquet)
|
| 8 |
dataset_info:
|
| 9 |
features:
|
|
|
|
|
|
|
| 10 |
- name: question
|
| 11 |
dtype: string
|
| 12 |
- name: question_type
|
|
@@ -16,8 +18,13 @@ dataset_info:
|
|
| 16 |
- name: image
|
| 17 |
sequence:
|
| 18 |
dtype: image
|
|
|
|
|
|
|
|
|
|
| 19 |
- name: id
|
| 20 |
dtype: string
|
|
|
|
|
|
|
| 21 |
- name: gt_value
|
| 22 |
dtype: bool
|
| 23 |
- name: relation
|
|
@@ -26,9 +33,6 @@ dataset_info:
|
|
| 26 |
dtype: string
|
| 27 |
- name: obj
|
| 28 |
dtype: string
|
| 29 |
-
- name: image_file
|
| 30 |
-
sequence:
|
| 31 |
-
dtype: string
|
| 32 |
splits:
|
| 33 |
- name: test
|
| 34 |
configs:
|
|
@@ -107,6 +111,4 @@ The `answer` is `"A"` if `gt_value` is `true`, otherwise `"B"`.
|
|
| 107 |
### Notes
|
| 108 |
- `<image>` placeholders are preserved in `question` and used to interleave images and text inside vlmevalkit prompts.
|
| 109 |
- Options (`A. True`, `B. False`) and the post prompt are embedded into `question`, so dataset consumers do not need to add choices externally.
|
| 110 |
-
- TSV uses base64-encoded images (string or JSON array string), while Parquet stores raw image bytes (`list[image]`).
|
| 111 |
-
|
| 112 |
-
|
|
|
|
| 7 |
pretty_name: VSR (Parquet)
|
| 8 |
dataset_info:
|
| 9 |
features:
|
| 10 |
+
- name: index
|
| 11 |
+
dtype: string
|
| 12 |
- name: question
|
| 13 |
dtype: string
|
| 14 |
- name: question_type
|
|
|
|
| 18 |
- name: image
|
| 19 |
sequence:
|
| 20 |
dtype: image
|
| 21 |
+
- name: image_file
|
| 22 |
+
sequence:
|
| 23 |
+
dtype: string
|
| 24 |
- name: id
|
| 25 |
dtype: string
|
| 26 |
+
- name: text
|
| 27 |
+
dtype: string
|
| 28 |
- name: gt_value
|
| 29 |
dtype: bool
|
| 30 |
- name: relation
|
|
|
|
| 33 |
dtype: string
|
| 34 |
- name: obj
|
| 35 |
dtype: string
|
|
|
|
|
|
|
|
|
|
| 36 |
splits:
|
| 37 |
- name: test
|
| 38 |
configs:
|
|
|
|
| 111 |
### Notes
|
| 112 |
- `<image>` placeholders are preserved in `question` and used to interleave images and text inside vlmevalkit prompts.
|
| 113 |
- Options (`A. True`, `B. False`) and the post prompt are embedded into `question`, so dataset consumers do not need to add choices externally.
|
| 114 |
+
- TSV uses base64-encoded images (string or JSON array string), while Parquet stores raw image bytes (`list[image]`).
|
|
|
|
|
|