lidingm commited on
Commit
715df9a
·
verified ·
1 Parent(s): fbe5147

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -1
README.md CHANGED
@@ -24,6 +24,7 @@ configs:
24
 
25
  <!-- Provide a longer summary of what this dataset is. -->
26
  We introduce ViewSpatial-Bench, a comprehensive benchmark with over 5,700 question-answer pairs across 1,000+ 3D scenes from ScanNet and MS-COCO validation sets. This benchmark evaluates VLMs' spatial localization capabilities from multiple perspectives, specifically testing both egocentric (camera) and allocentric (human subject) viewpoints across five distinct task types.
 
27
  ViewSpatial-Bench addresses a critical gap: while VLMs excel at spatial reasoning from their own perspective, they struggle with perspective-taking—adopting another entity's spatial frame of reference—which is essential for embodied interaction and multi-agent collaboration.
28
  - **Language(s) (NLP):** en
29
  - **License:** apache-2.0
@@ -78,7 +79,7 @@ We provide benchmark results for various open-source models as well as **GPT-4o*
78
  <td>34.98</td>
79
  </tr>
80
  <tr>
81
- <td>Qwen2.5-VL (3B) [Backbone]</td>
82
  <td>43.43</td><td>33.33</td><td>39.80</td>
83
  <td>39.16</td><td>28.62</td><td>28.51</td><td>32.14</td>
84
  <td>35.85</td>
 
24
 
25
  <!-- Provide a longer summary of what this dataset is. -->
26
  We introduce ViewSpatial-Bench, a comprehensive benchmark with over 5,700 question-answer pairs across 1,000+ 3D scenes from ScanNet and MS-COCO validation sets. This benchmark evaluates VLMs' spatial localization capabilities from multiple perspectives, specifically testing both egocentric (camera) and allocentric (human subject) viewpoints across five distinct task types.
27
+
28
  ViewSpatial-Bench addresses a critical gap: while VLMs excel at spatial reasoning from their own perspective, they struggle with perspective-taking—adopting another entity's spatial frame of reference—which is essential for embodied interaction and multi-agent collaboration.
29
  - **Language(s) (NLP):** en
30
  - **License:** apache-2.0
 
79
  <td>34.98</td>
80
  </tr>
81
  <tr>
82
+ <td>Qwen2.5-VL (3B)</td>
83
  <td>43.43</td><td>33.33</td><td>39.80</td>
84
  <td>39.16</td><td>28.62</td><td>28.51</td><td>32.14</td>
85
  <td>35.85</td>