kd7 nielsr HF Staff commited on
Commit
304ff11
·
verified ·
1 Parent(s): 3254372

Enhance dataset card: Add paper/project/code links, abstract, paper citation, and update task categories (#1)

Browse files

- Enhance dataset card: Add paper/project/code links, abstract, paper citation, and update task categories (fc21fdb8bb13d0d5b39337a01220b36f7af1cb0c)


Co-authored-by: Niels Rogge <[email protected]>

Files changed (1) hide show
  1. README.md +17 -4
README.md CHANGED
@@ -1,11 +1,12 @@
1
  ---
2
- pretty_name: GRAID Waymo Perception Dataset Question-Answer Dataset
3
  language:
4
  - en
5
  license: cc-by-nc-4.0
6
  task_categories:
7
  - visual-question-answering
8
  - object-detection
 
 
9
  tags:
10
  - visual-reasoning
11
  - spatial-reasoning
@@ -61,12 +62,17 @@ configs:
61
 
62
  # GRAID Waymo Perception Dataset Question-Answer Dataset
63
 
 
 
64
  ## Overview
65
 
66
  This dataset was generated using **GRAID** (**G**enerating **R**easoning questions from **A**nalysis of **I**mages via **D**iscriminative artificial intelligence), a framework for creating spatial reasoning datasets from object detection annotations.
67
 
68
  **GRAID** transforms raw object detection data into structured question-answer pairs that test various aspects of object localization, visual reasoning, spatial reasoning, and object relationship comprehension.
69
 
 
 
 
70
  ## Dataset Details
71
 
72
  - **Total QA Pairs**: 13,855
@@ -142,14 +148,14 @@ train_data = dataset["train"]
142
  val_data = dataset["val"]
143
 
144
  # Example of accessing a sample
145
- sample = dataset["train"][0] # or "val"
146
  print(f"Question: {sample['question']}")
147
  print(f"Answer: {sample['answer']}")
148
  print(f"Question Type: {sample['question_type']}")
149
 
150
  # The image is embedded as a PIL Image object
151
  image = sample["image"]
152
- image.show() # Display the image
153
  ```
154
 
155
  ## Dataset Schema
@@ -179,6 +185,13 @@ This dataset is derived from the Waymo Perception Dataset. Please refer to the [
179
  If you use this dataset in your research, please cite both the original dataset and the GRAID framework:
180
 
181
  ```bibtex
 
 
 
 
 
 
 
182
  @dataset{graid_waymo,
183
  title={GRAID Waymo Perception Dataset Question-Answer Dataset},
184
  author={GRAID Framework},
@@ -197,4 +210,4 @@ If you use this dataset in your research, please cite both the original dataset
197
 
198
  ## Contact
199
 
200
- For questions about this dataset or the GRAID framework, please open an issue in the repository.
 
1
  ---
 
2
  language:
3
  - en
4
  license: cc-by-nc-4.0
5
  task_categories:
6
  - visual-question-answering
7
  - object-detection
8
+ - image-text-to-text
9
+ pretty_name: GRAID Waymo Perception Dataset Question-Answer Dataset
10
  tags:
11
  - visual-reasoning
12
  - spatial-reasoning
 
62
 
63
  # GRAID Waymo Perception Dataset Question-Answer Dataset
64
 
65
+ [Paper](https://huggingface.co/papers/2510.22118) | [Project Page](https://ke7.github.io/graid/) | [Code](https://github.com/kd7-ml/graid)
66
+
67
  ## Overview
68
 
69
  This dataset was generated using **GRAID** (**G**enerating **R**easoning questions from **A**nalysis of **I**mages via **D**iscriminative artificial intelligence), a framework for creating spatial reasoning datasets from object detection annotations.
70
 
71
  **GRAID** transforms raw object detection data into structured question-answer pairs that test various aspects of object localization, visual reasoning, spatial reasoning, and object relationship comprehension.
72
 
73
+ ## Abstract
74
+ Vision Language Models (VLMs) achieve strong performance on many vision-language tasks but often struggle with spatial reasoning\textemdash{}a prerequisite for many applications. Empirically, we find that a dataset produced by a current training data generation pipeline has a 57.6\% human validation rate. These rates stem from current limitations: single-image 3D reconstruction introduces cascading modeling errors and requires wide answer tolerances, while caption-based methods require hyper-detailed annotations and suffer from generative hallucinations. We present GRAID, built on the key insight that qualitative spatial relationships can be reliably determined from 2D geometric primitives alone. By operating exclusively on 2D bounding boxes from standard object detectors, GRAID avoids both 3D reconstruction errors and generative hallucinations, resulting in datasets that are of higher quality than existing tools that produce similar datasets as validated by human evaluations. We apply our framework to the BDD100k, NuImages, and Waymo datasets, generating over 8.5 million high-quality VQA pairs creating questions spanning spatial relations, counting, ranking, and size comparisons. We evaluate one of the datasets and find it achieves 91.16\% human-validated accuracy\textemdash{}compared to 57.6\% on a dataset generated by recent work. % or recent work Critically, we demonstrate that when trained on GRAID data, models learn spatial reasoning concepts that generalize: models fine-tuned on 6 question types improve on over 10 held-out types, with accuracy gains of 47.5\% on BDD and 37.9\% on NuImages for Llama 3.2B 11B, and when trained on all questions types, achieve improvements on several existing benchmarks such as BLINK. The GRAID framework, datasets, and additional information can be found on our \href{ this https URL }{project page}.
75
+
76
  ## Dataset Details
77
 
78
  - **Total QA Pairs**: 13,855
 
148
  val_data = dataset["val"]
149
 
150
  # Example of accessing a sample
151
+ sample = dataset["train"][0] # or "val"
152
  print(f"Question: {sample['question']}")
153
  print(f"Answer: {sample['answer']}")
154
  print(f"Question Type: {sample['question_type']}")
155
 
156
  # The image is embedded as a PIL Image object
157
  image = sample["image"]
158
+ image.show() # Display the image
159
  ```
160
 
161
  ## Dataset Schema
 
185
  If you use this dataset in your research, please cite both the original dataset and the GRAID framework:
186
 
187
  ```bibtex
188
+ @article{elmaaroufi2025graid,
189
+ title={GRAID: Enhancing Spatial Reasoning of VLMs Through High-Fidelity Data Generation},
190
+ author={Elmaaroufi, Karim and Zheng, Jonathan and Xu, Haohan and Pan, Yang and Kim, Younghyun and Choi, Joshua and Amsalem, Yoav and Ma, Jianxin and Xu, Minjun and Liu, Fang and Liang, Ting and Singh, Kavit and Hwu, Wen-mei and Chen, Yida},
191
+ journal={arXiv preprint arXiv:2510.22118},
192
+ year={2025}
193
+ }
194
+
195
  @dataset{graid_waymo,
196
  title={GRAID Waymo Perception Dataset Question-Answer Dataset},
197
  author={GRAID Framework},
 
210
 
211
  ## Contact
212
 
213
+ For questions about this dataset or the GRAID framework, please open an issue in the repository.