py2279105943 commited on
Commit
89563c9
·
verified ·
1 Parent(s): bf15066

Add per-video intent metadata for Dataset Viewer

Browse files
Files changed (2) hide show
  1. README.md +16 -0
  2. metadata.jsonl +0 -0
README.md CHANGED
@@ -15,6 +15,9 @@ tags:
15
  - procedural-reasoning
16
  - action-anticipation
17
  - multimodal
 
 
 
18
  ---
19
 
20
  # EgoIntent
@@ -59,6 +62,7 @@ The three targets are related but non-redundant. Because valid descriptions may
59
 
60
  ```text
61
  EgoIntent/
 
62
  ├── indoor/
63
  │ └── art_studio/
64
  │ └── draw/
@@ -84,6 +88,18 @@ Paths follow:
84
 
85
  where `setting` is `indoor` or `outdoor`.
86
 
 
 
 
 
 
 
 
 
 
 
 
 
87
  ## Annotation Format
88
 
89
  Each event directory contains one `step_label.json` file:
 
15
  - procedural-reasoning
16
  - action-anticipation
17
  - multimodal
18
+ configs:
19
+ - config_name: default
20
+ drop_labels: true
21
  ---
22
 
23
  # EgoIntent
 
62
 
63
  ```text
64
  EgoIntent/
65
+ ├── metadata.jsonl
66
  ├── indoor/
67
  │ └── art_studio/
68
  │ └── draw/
 
88
 
89
  where `setting` is `indoor` or `outdoor`.
90
 
91
+ ### Dataset Viewer Metadata
92
+
93
+ The root-level `metadata.jsonl` contains one row per MP4 and links the video through its relative `file_name`. Hugging Face renders that field as the `video` column. The remaining Viewer columns are:
94
+
95
+ - `id`: a globally unique ID in the form `{setting}/{scene}/{event}/step{step_id}`;
96
+ - `local_intent`;
97
+ - `procedural_intent`;
98
+ - `observed_next_step`;
99
+ - `plausible_next_steps`.
100
+
101
+ The Dataset Card configuration explicitly sets `drop_labels: true`, so event directory names are not interpreted as classification labels.
102
+
103
  ## Annotation Format
104
 
105
  Each event directory contains one `step_label.json` file:
metadata.jsonl ADDED
The diff for this file is too large to render. See raw diff