Commit
•
8320a57
1
Parent(s):
20fa2f1
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,15 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
CRPE is a benchmark designed to quantitatively evaluate the object recognition and relation comprehension ability of models.
|
| 6 |
+
The evaluation is formulated as single-choice questions.
|
| 7 |
+
|
| 8 |
+
The benchmark consists of four splits:
|
| 9 |
+
**Existence**, **Subject**, **Predicate**, and **Object**.
|
| 10 |
+
The **Existence** split evaluates the object recognition ability while the remaining splits are designed to evaluate the capability of relation comprehension, focusing on probing each of the elements in the subject-predicate-object triplets of the scene graph separately.
|
| 11 |
+
Some data examples are shown below.
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
See our [paper](https://github.com/OpenGVLab/all-seeing/all-seeing-v2) to learn more details!
|