Simplify model evaluation introduction
Browse files- README.md +2 -2
- README_zh-CN.md +1 -1
README.md
CHANGED
|
@@ -44,8 +44,8 @@ Lei Zhang<sup>1,2</sup>
|
|
| 44 |
|
| 45 |
We introduce **OmniCompass**, a holistic benchmark for omni-modal video grounding
|
| 46 |
and understanding with paired Text/Text+Image target specifications and
|
| 47 |
-
complementary visual and audio-visual tracks.
|
| 48 |
-
|
| 49 |
answering, assessing both task performance and cross-reference consistency.
|
| 50 |
|
| 51 |
## 🔥 News
|
|
|
|
| 44 |
|
| 45 |
We introduce **OmniCompass**, a holistic benchmark for omni-modal video grounding
|
| 46 |
and understanding with paired Text/Text+Image target specifications and
|
| 47 |
+
complementary visual and audio-visual tracks.
|
| 48 |
+
We evaluate **50 MLLMs** on temporal grounding, spatial grounding, and question
|
| 49 |
answering, assessing both task performance and cross-reference consistency.
|
| 50 |
|
| 51 |
## 🔥 News
|
README_zh-CN.md
CHANGED
|
@@ -32,7 +32,7 @@ Lei Zhang<sup>1,2</sup>
|
|
| 32 |
|
| 33 |
我们提出 **OmniCompass**,一个面向全模态视频定位与理解的综合评测基准,
|
| 34 |
包含成对的 Text/Text+Image 目标描述,以及相互补充的视觉和音视频评测轨道。
|
| 35 |
-
我们
|
| 36 |
同时衡量任务表现与跨参考形式的一致性。
|
| 37 |
|
| 38 |
## 🔥 最新动态
|
|
|
|
| 32 |
|
| 33 |
我们提出 **OmniCompass**,一个面向全模态视频定位与理解的综合评测基准,
|
| 34 |
包含成对的 Text/Text+Image 目标描述,以及相互补充的视觉和音视频评测轨道。
|
| 35 |
+
我们评测了 **50 个 MLLMs**,覆盖时间定位、空间定位和问答,
|
| 36 |
同时衡量任务表现与跨参考形式的一致性。
|
| 37 |
|
| 38 |
## 🔥 最新动态
|