Datasets:

ArXiv:
wushengqiong commited on
Commit
fa15ad2
Β·
1 Parent(s): f541f38

update readme

Browse files
Files changed (1) hide show
  1. README.md +126 -123
README.md CHANGED
@@ -36,15 +36,10 @@ Please download the dataset accordingly.
36
 
37
  ## πŸ“• Table of Contents
38
 
39
- - [✨ File Origanization Structure](#filestructure)
40
- - [🍟 Usage](#usage)
41
- - [🌐 General-Bench](#bench)
42
- - [πŸ• Capabilities and Domians Distribution](#distribution)
43
- - [πŸ–ΌοΈ Image Task Taxonomy](#imagetaxonomy)
44
- - [πŸ“½οΈ Video Task Taxonomy](#videotaxonomy)
45
- - [πŸ“ž Audio Task Taxonomy](#audiotaxonomy)
46
- - [πŸ’Ž 3D Task Taxonomy](#3dtaxonomy)
47
- - [πŸ“š Language Task Taxonomy](#languagetaxonomy)
48
 
49
 
50
 
@@ -52,68 +47,22 @@ Please download the dataset accordingly.
52
 
53
  ---
54
 
55
- <span id='filestructure'/>
56
 
57
- # ✨✨✨ **File Origanization Structure**
58
 
59
- Here is the organization structure of the file system:
 
 
 
 
 
 
 
 
60
 
61
  ```
62
- General-Bench
63
- β”œβ”€β”€ Image
64
- β”‚ β”œβ”€β”€ comprehension
65
- β”‚ β”‚ β”œβ”€β”€ Bird-Detection
66
- β”‚ β”‚ β”‚ β”œβ”€β”€ annotation.json
67
- β”‚ β”‚ β”‚ └── images
68
- β”‚ β”‚ β”‚ └── Acadian_Flycatcher_0070_29150.jpg
69
- β”‚ β”‚ β”œβ”€β”€ Bottle-Anomaly-Detection
70
- β”‚ β”‚ β”‚ β”œβ”€β”€ annotation.json
71
- β”‚ β”‚ β”‚ └── images
72
- β”‚ β”‚ └── ...
73
- β”‚ └── generation
74
- β”‚ └── Layout-to-Face-Image-Generation
75
- β”‚ β”œβ”€β”€ annotation.json
76
- β”‚ └── images
77
- β”‚ └── ...
78
- β”œβ”€β”€ Video
79
- β”‚ β”œβ”€β”€ comprehension
80
- β”‚ β”‚ └── Human-Object-Interaction-Video-Captioning
81
- β”‚ β”‚ β”œβ”€β”€ annotation.json
82
- β”‚ β”‚ └── videos
83
- β”‚ β”‚ └── ...
84
- β”‚ └── generation
85
- β”‚ └── Scene-Image-to-Video-Generation
86
- β”‚ β”œβ”€β”€ annotation.json
87
- β”‚ └── videos
88
- β”‚ └── ...
89
- β”œβ”€β”€ 3d
90
- β”‚ β”œβ”€β”€ comprehension
91
- β”‚ β”‚ └── 3D-Furniture-Classification
92
- β”‚ β”‚ β”œβ”€β”€ annotation.json
93
- β”‚ β”‚ └── pointclouds
94
- β”‚ β”‚ └── ...
95
- β”‚ └── generation
96
- β”‚ └── Text-to-3D-Living-and-Arts-Point-Cloud-Generation
97
- β”‚ β”œβ”€β”€ annotation.json
98
- β”‚ └── pointclouds
99
- β”‚ └── ...
100
- β”œβ”€β”€ Audio
101
- β”‚ β”œβ”€β”€ comprehension
102
- β”‚ β”‚ └── Accent-Classification
103
- β”‚ β”‚ β”œβ”€β”€ annotation.json
104
- β”‚ β”‚ └── audios
105
- β”‚ β”‚ └── ...
106
- β”‚ └── generation
107
- β”‚ └── Video-To-Audio
108
- β”‚ β”œβ”€β”€ annotation.json
109
- β”‚ └── audios
110
- β”‚ └── ...
111
- β”œβ”€β”€ NLP
112
- β”‚ β”œβ”€β”€ History-Question-Answering
113
- β”‚ β”‚ └── annotation.json
114
- β”‚ β”œβ”€β”€ Abstractive-Summarization
115
- β”‚ β”‚ └── annotation.json
116
- β”‚ └── ...
117
 
118
  ```
119
 
@@ -124,13 +73,6 @@ An illustrative example of file formats:
124
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/RD3b7Jwu0dftVq-4KbpFr.png)
125
 
126
 
127
- <span id='usage'/>
128
-
129
- ## 🍟🍟🍟 Usage
130
-
131
- Please download all the files in this repository. We also provide overview.json, which is an example of the format of our dataset.
132
-
133
- For more instructions, please go to the [document page](https://generalist.top/document).
134
 
135
 
136
  ---
@@ -139,105 +81,166 @@ For more instructions, please go to the [document page](https://generalist.top/d
139
 
140
 
141
 
142
- <span id='bench'/>
143
 
144
 
145
 
146
- # 🌐🌐🌐 **General-Bench**
147
 
148
 
149
 
150
 
151
- A companion massive multimodal benchmark dataset, encompasses a broader spectrum of skills, modalities, formats, and capabilities, including over **`700`** tasks and **`325K`** instances.
152
-
153
- <div align="center">
154
- <img src='https://cdn-uploads.huggingface.co/production/uploads/647773a1168cb428e00e9a8f/d4TIWw3rlWuxpBCEpHYJB.jpeg'>
155
- <p> Overview of General-Bench, which covers 145 skills for more than 700 tasks with over 325,800 samples under
156
- comprehension and generation categories in various modalities</p>
157
- </div>
158
 
 
 
 
159
 
160
- <span id='distribution'/>
161
 
162
- ## πŸ•πŸ•πŸ• Capabilities and Domians Distribution
 
 
163
 
164
- <div align="center">
165
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/fF3iH95B3QEBvJYwqzZVG.png'>
166
- <p> Distribution of various capabilities evaluated in General-Bench.</p>
167
- </div>
168
 
 
 
169
 
170
- <div align="center">
171
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/wQvllVeK-KC3Edp8Zjh-V.png'>
172
- <p>Distribution of various domains and disciplines covered by General-Bench.</p>
173
- </div>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
174
 
 
175
 
176
 
177
 
 
178
 
179
- <span id='imagetaxonomy'/>
180
 
181
- # πŸ–ΌοΈ Image Task Taxonomy
182
- <div align="center">
183
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/2QYihQRhZ5C9K5IbukY7R.png'>
184
- <p>Taxonomy and hierarchy of data in terms of Image modality.</p>
185
- </div>
186
 
187
 
188
 
 
189
 
190
- <span id='videotaxonomy'/>
191
 
192
- # πŸ“½οΈ Video Task Taxonomy
193
 
194
- <div align="center">
195
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/A7PwfW5gXzstkDH49yIG5.png'>
196
- <p>Taxonomy and hierarchy of data in terms of Video modality.</p>
197
- </div>
198
 
199
 
 
 
 
200
 
201
 
202
 
203
 
 
 
204
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
205
 
 
206
 
207
- <span id='audiotaxonomy'/>
208
 
209
- # πŸ“ž Audio Task Taxonomy
210
 
 
211
 
 
212
 
213
- <div align="center">
214
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/e-QBvBjeZy8vmcBjAB0PE.png'>
215
- <p>Taxonomy and hierarchy of data in terms of Audio modality.</p>
216
- </div>
217
 
 
 
 
218
 
219
 
220
- <span id='3dtaxonomy'/>
 
 
 
221
 
222
- # πŸ’Ž 3D Task Taxonomy
223
 
224
 
225
- <div align="center">
226
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/EBXb-wyve14ExoLCgrpDK.png'>
227
- <p>Taxonomy and hierarchy of data in terms of 3D modality.</p>
228
- </div>
229
 
230
 
 
 
 
 
231
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
232
 
233
- <span id='languagetaxonomy'/>
234
 
235
- # πŸ“š Language Task Taxonomy
236
 
237
- <div align="center">
238
- <img src='https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/FLfk3QGdYb2sgorKTj_LT.png'>
239
- <p>Taxonomy and hierarchy of data in terms of Language modality.</p>
240
- </div>
241
 
242
 
243
 
 
36
 
37
  ## πŸ“• Table of Contents
38
 
39
+ - [✨ Scope-A](#scope_a)
40
+ - [🌐 Scope-B](#scope_b)
41
+ - [πŸ–ΌοΈ Scope-C](#scope_c)
42
+ - [πŸ“½οΈ Scope-D](#scope_d)
 
 
 
 
 
43
 
44
 
45
 
 
47
 
48
  ---
49
 
50
+ <span id='scope_a'/>
51
 
52
+ # ✨✨✨ **Scope-A**
53
 
54
+ Full-spectrum leaderboard covering all modalities and tasks under General-Level, for highly capable, general-purpose multimodal models.
55
+
56
+ - πŸ” Details:
57
+ - βœ”οΈ Covers all General-Level tasks and modalities.
58
+ - βœ”οΈ Most challenging track; requires high model capacity and resource commitment.
59
+
60
+ - πŸŽ‰ Highlights:
61
+ - βœ”οΈ Evaluates holistic generalization and cross-modal synergy.
62
+ - βœ”οΈ Suitable for near-AGI or foundation-level multimodal generalists.
63
 
64
  ```
65
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
66
 
67
  ```
68
 
 
73
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/64c139d867eff857ea51caa8/RD3b7Jwu0dftVq-4KbpFr.png)
74
 
75
 
 
 
 
 
 
 
 
76
 
77
 
78
  ---
 
81
 
82
 
83
 
84
+ <span id='scope_b'/>
85
 
86
 
87
 
88
+ # 🌐🌐🌐 Scope-B
89
 
90
 
91
 
92
 
93
+ Modality-specific leaderboards focusing on single modality or partially joint modality (e.g., image, video, audio, 3D) for modality-wise generalists.
 
 
 
 
 
 
94
 
95
+ - πŸ” Details:
96
+ - βœ”οΈ 7 separate leaderboards (4 single modality + 3 combined modality).
97
+ - βœ”οΈ Focuses on mastering diverse tasks within a single modality.
98
 
 
99
 
100
+ - πŸŽ‰ Highlights:
101
+ - βœ”οΈ Measures within-modality generalization.
102
+ - βœ”οΈ Suited for intermediate-level models with cross-task transferability.
103
 
 
 
 
 
104
 
105
+ In [Scope-B](https://huggingface.co/datasets/General-Level/General-Bench-Closeset-Scoped/tree/main/Scope-B), we provide the subset of data corresponding to each sub-leaderboard. Each task is represented by a separate JSON file, which specifies the dataset associated with that particular sub-leaderboard, including the relevant file names.
106
+ All referenced data files can be found in the [πŸ‘‰ **`Close Set`**](https://huggingface.co/datasets/General-Level/General-Bench-Closeset) repository.
107
 
108
+ ```
109
+ {
110
+ ## paradigm
111
+ "comprehension": {
112
+ ## skill name
113
+ "Speech Accent Understanding":
114
+ [
115
+ {
116
+ ## task name:data file name in Closeset
117
+ "Accent Classification": "AccentClassification"
118
+ },
119
+ {
120
+ "Accent Sex Classification": "AccentSexClassification"
121
+ },
122
+ {
123
+ "Speaker Identification": "SpeakerIdentification"
124
+ },
125
+ {
126
+ "Vocal Sound Classification": "VocalSoundClassification"
127
+ }
128
+ ],
129
+ ...
130
+ }
131
+ }
132
 
133
+ ```
134
 
135
 
136
 
137
+ ----
138
 
 
139
 
 
 
 
 
 
140
 
141
 
142
 
143
+ <span id='scope_c'/>
144
 
145
+ # πŸ–ΌοΈπŸ–ΌοΈπŸ–ΌοΈ Scope-C
146
 
147
+ Leaderboards categorized by comprehension vs. generation tasks within each modality. Lower entry barrier for early-stage or lightweight models.
148
 
149
+ - πŸ” Details:
150
+ - βœ”οΈ 8 leaderboards: 2 Γ— 4 for multimodal comprehension/generation under different modalities.
151
+ - βœ”οΈ Supports entry-level model evaluation or teams with limited resources.
 
152
 
153
 
154
+ - πŸŽ‰ Highlights:
155
+ - βœ”οΈ Assesses task-type specialization: understanding or generation.
156
+ - βœ”οΈ Reflects generalization across task types.
157
 
158
 
159
 
160
 
161
+ In [Scope-C](https://huggingface.co/datasets/General-Level/General-Bench-Closeset-Scoped/tree/main/Scope-C), we provide the subset of data corresponding to each sub-leaderboard. Each task is represented by a separate JSON file, which specifies the dataset associated with that particular sub-leaderboard, including the relevant file names.
162
+ All referenced data files can be found in the [πŸ‘‰ **`Close Set`**](https://huggingface.co/datasets/General-Level/General-Bench-Closeset) repository.
163
 
164
+ ```
165
+ {
166
+ ## skill name
167
+ "3D Human-related Object Classification": [
168
+ {
169
+ ## task name:data file name in Closeset
170
+ "3D Accessory Classification": "3d_classification/ModelNet40/accessory"
171
+ },
172
+ {
173
+ "3D Appliance Classification": "3d_classification/ModelNet40/appliance"
174
+ },
175
+ {
176
+ "3D Tableware Classification": "3d_classification/ModelNet40/tableware"
177
+ },
178
+ {
179
+ "3D Musical Instrument Classification": "3d_classification/ModelNet40/musical_instrument"
180
+ },
181
+ {
182
+ "3D Person Classification": "3d_classification/ModelNet40/person"
183
+ }
184
+ ],
185
+ ...
186
+ }
187
 
188
+ ```
189
 
190
+ ----
191
 
 
192
 
193
+ <span id='scope_d'/>
194
 
195
+ # πŸ“½οΈπŸ“½οΈπŸ“½οΈ Scope-D
196
 
197
+ Fine-grained leaderboards focused on specific task clusters (e.g., VQA, Captioning, Speech Recognition), ideal for partial generalists.
 
 
 
198
 
199
+ - πŸ” Details:
200
+ - βœ”οΈ Large number of sub-leaderboards, each scoped to a skill set.
201
+ - βœ”οΈ Easiest to participate; lowest cost.
202
 
203
 
204
+ - πŸŽ‰ Highlights:
205
+ - βœ”οΈ Evaluates fine-grained skill performance.
206
+ - βœ”οΈ Helps identify model strengths and specialization areas.
207
+ - βœ”οΈ Encourages progressive development toward broader leaderboard participation.
208
 
 
209
 
210
 
 
 
 
 
211
 
212
 
213
+ In [Scope-D](https://huggingface.co/datasets/General-Level/General-Bench-Closeset-Scoped/tree/main/Scope-D), we provide subset datasets corresponding to each sub-leaderboard. Each task is represented by a JSON file named in the format:
214
+ `{modality name}β€”β€”{comp/gen}_{clustered skill name}.json`.
215
+ Each JSON file specifies the dataset used for the corresponding sub-leaderboard task, including the list of relevant file names.
216
+ All referenced data files can be found in the [πŸ‘‰ **`Close Set`**](https://huggingface.co/datasets/General-Level/General-Bench-Closeset) repository.
217
 
218
+ ```
219
+ {
220
+ ## clusered skill name
221
+ "Classifcation": {
222
+ ## skill name
223
+ "3D Human-related Object Classification": [
224
+ "3d_classification/ModelNet40/accessory", ## data file name in Closeset
225
+ "3d_classification/ModelNet40/appliance",
226
+ "3d_classification/ModelNet40/tableware",
227
+ "3d_classification/ModelNet40/musical_instrument",
228
+ "3d_classification/ModelNet40/person"
229
+ ],
230
+ "3D Structure and Environment Classification": [
231
+ "3d_classification/ModelNet40/furniture",
232
+ "3d_classification/ModelNet40/structure"
233
+ ],
234
+ "Transportation and Technology Object Classification": [
235
+ "3d_classification/ModelNet40/electronic",
236
+ "3d_classification/ModelNet40/vehicle"
237
+ ]
238
+ }
239
+ }
240
 
241
+ ```
242
 
 
243
 
 
 
 
 
244
 
245
 
246