Update README.md
Browse files
README.md
CHANGED
|
@@ -158,69 +158,109 @@ configs:
|
|
| 158 |
---
|
| 159 |
|
| 160 |
|
| 161 |
-
<
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
|
| 163 |
|
| 164 |
The RoomSpace dataset, associated with the IJCAI-24 paper titled "Reframing Spatial Reasoning Evaluation in Language Models: A Real-World Simulation Benchmark for Qualitative Reasoning," aims to enhance the evaluation of spatial reasoning in language models. This benchmark encompasses a comprehensive range of qualitative spatial relationships, including topological, directional, and distance relations. These relationships are presented from various viewpoints, with differing levels of granularity and density of relational constraints to simulate real-world complexities. This approach promotes a more accurate assessment of language models' capabilities in spatial reasoning tasks.
|
| 165 |
|
| 166 |
---
|
|
|
|
| 167 |
|
| 168 |
-
##
|
| 169 |
|
| 170 |
-
|
| 171 |
|
| 172 |
```
|
| 173 |
Data/
|
| 174 |
-
βββ
|
| 175 |
-
βββ SD-
|
| 176 |
-
|
|
|
|
| 177 |
```
|
| 178 |
|
| 179 |
-
- `SD-
|
| 180 |
-
- `SD-
|
| 181 |
-
- `SD-
|
|
|
|
|
|
|
| 182 |
|
| 183 |
-
|
| 184 |
|
| 185 |
```
|
| 186 |
-
|
| 187 |
-
βββ
|
| 188 |
-
βββ
|
| 189 |
-
β βββ top-down/
|
| 190 |
-
β βββ robot_south/
|
| 191 |
-
β βββ robot_south_rec/
|
| 192 |
-
βββ Text/
|
| 193 |
-
βββ Logic/
|
| 194 |
```
|
| 195 |
|
| 196 |
-
|
|
|
|
|
|
|
| 197 |
|
| 198 |
-
|
| 199 |
|
| 200 |
-
|
|
|
|
| 201 |
|
| 202 |
-
|
| 203 |
|
| 204 |
-
|
| 205 |
-
- `robot_south`: North-facing square images, assuming a robot stands at the south door.
|
| 206 |
-
- `robot_south_rec`: North-facing rectangular images.
|
| 207 |
|
| 208 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 209 |
|
| 210 |
-
###
|
| 211 |
|
| 212 |
-
|
|
|
|
| 213 |
|
| 214 |
-
###
|
| 215 |
|
| 216 |
-
|
|
|
|
| 217 |
|
| 218 |
-
- `n*_m*_d*.json`: Based on `n`, `m`, and `d` parameters as described.
|
| 219 |
-
- `answers_lengths_d*.json`: Records the answer lengths for each setting and each example for different `d` values.
|
| 220 |
-
- `times_fr_take_d*.json`: CPU time taken for each setting of the find relation question.
|
| 221 |
-
- `times_yn_take_d*.json`: CPU time taken for each setting of the yes/no question.
|
| 222 |
-
- `skip_id_d*.json`: Records the example IDs that do not meet the `m` requirements for the object numbers, which will be skipped.
|
| 223 |
-
- `solution_id_d*.json`: Records the IDs that have solutions (answer length > 0) under each spatial setting.
|
| 224 |
|
| 225 |
---
|
| 226 |
## Relevant Paper
|
|
|
|
| 158 |
---
|
| 159 |
|
| 160 |
|
| 161 |
+
<div align="center">
|
| 162 |
+
<h1>π RoomSpace Benchmark</h1>
|
| 163 |
+
<p>
|
| 164 |
+
π <a href="https://roomspace-benchmark.web.app/">Homepage</a> β’
|
| 165 |
+
π <a href="https://github.com/Fangjun-Li/RoomSpace">GitHub</a> β’
|
| 166 |
+
π <a href="https://archive.researchdata.leeds.ac.uk/1293/">Dataset</a> β’
|
| 167 |
+
π <a href="https://arxiv.org/abs/2405.15064">Paper</a>
|
| 168 |
+
</p>
|
| 169 |
+
<p><em>Simplify LLM evaluation using a convenient Colab notebook.</em></p>
|
| 170 |
+
<a href="https://colab.research.google.com/drive/1fAK8J1UHAjMm-mNVsuzIbEZd-SZG6bX-?usp=sharing"><img src="img/colab.svg" alt="Open In Colab"></a></center>
|
| 171 |
+
</div>
|
| 172 |
+
<br/>
|
| 173 |
|
| 174 |
|
| 175 |
The RoomSpace dataset, associated with the IJCAI-24 paper titled "Reframing Spatial Reasoning Evaluation in Language Models: A Real-World Simulation Benchmark for Qualitative Reasoning," aims to enhance the evaluation of spatial reasoning in language models. This benchmark encompasses a comprehensive range of qualitative spatial relationships, including topological, directional, and distance relations. These relationships are presented from various viewpoints, with differing levels of granularity and density of relational constraints to simulate real-world complexities. This approach promotes a more accurate assessment of language models' capabilities in spatial reasoning tasks.
|
| 176 |
|
| 177 |
---
|
| 178 |
+
# RoomSpace Dataset Repository
|
| 179 |
|
| 180 |
+
## Branch Structure
|
| 181 |
|
| 182 |
+
This repository contains various branches, each representing different versions and sizes of the RoomSpace dataset. For example, the `SD-100_v1.0` branch includes 100 test examples under version 1.0.
|
| 183 |
|
| 184 |
```
|
| 185 |
Data/
|
| 186 |
+
βββ main/
|
| 187 |
+
βββ SD-100_v1.0/
|
| 188 |
+
βββ SD-1K_v1.0/
|
| 189 |
+
βββ SD-10K_v1.0/
|
| 190 |
```
|
| 191 |
|
| 192 |
+
- `SD-100_v1.0`: Contains data for 100 room samples.
|
| 193 |
+
- `SD-1K_v1.0`: Contains data for 1,000 room samples, with the first 100 rooms identical to those in `SD-100_v1.0`.
|
| 194 |
+
- `SD-10K_v1.0`: Contains data for 10,000 room samples, with the first 1,000 rooms identical to those in `SD-1K_v1.0`.
|
| 195 |
+
|
| 196 |
+
## Directory Structure
|
| 197 |
|
| 198 |
+
The repository is organized as follows:
|
| 199 |
|
| 200 |
```
|
| 201 |
+
Revision/
|
| 202 |
+
βββ n3_m2_d144/
|
| 203 |
+
βββ n4_m3_d144/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 204 |
```
|
| 205 |
|
| 206 |
+
Each `n*_m*_d*` subdirectory includes a test parquet file containing stories, questions, and answers derived from the images and a Constraint Satisfaction Problem (CSP) reasoner. The parameters `n`, `m`, and `d` represent the object number, constraint density, and domain size, respectively.
|
| 207 |
+
|
| 208 |
+
### Images
|
| 209 |
|
| 210 |
+
Each example includes two images showing different views of the rooms:
|
| 211 |
|
| 212 |
+
- `image_top_down`: Top-down images of the rooms.
|
| 213 |
+
- `image_north_facing`: North-facing rectangular images, assuming a robot stands at the south door, facing north.
|
| 214 |
|
| 215 |
+
### Questions & Answers
|
| 216 |
|
| 217 |
+
There are a total of 26 QA pairs, categorized as follows:
|
|
|
|
|
|
|
| 218 |
|
| 219 |
+
- **Visual QA**:
|
| 220 |
+
- (image_top_down + question_image) -- answer_image_td
|
| 221 |
+
- (image_north_facing + question_image) -- answer_image_sd
|
| 222 |
+
- **Layout Setting**:
|
| 223 |
+
- (question_td_layout_yn -- answer_td_layout_yn)
|
| 224 |
+
- (question_td_layout_fr -- answer_td_layout_fr)
|
| 225 |
+
- **TPP Setting**:
|
| 226 |
+
- (question_td_tpp_yn -- answer_td_tpp_yn)
|
| 227 |
+
- (question_td_tpp_fr -- answer_td_tpp_fr)
|
| 228 |
+
- **O2 Setting**:
|
| 229 |
+
- (question_td_o2_yn -- answer_td_o2_yn)
|
| 230 |
+
- (question_td_o2_fr -- answer_td_o2_fr)
|
| 231 |
+
- (question_sd_o2_yn -- answer_sd_o2_yn)
|
| 232 |
+
- (question_sd_o2_fr -- answer_sd_o2_fr)
|
| 233 |
+
- **O2+D2 Setting**:
|
| 234 |
+
- (question_td_o2_d2_yn -- answer_td_o2_d2_yn)
|
| 235 |
+
- (question_td_o2_d2_fr -- answer_td_o2_d2_fr)
|
| 236 |
+
- (question_sd_o2_d2_yn -- answer_sd_o2_d2_yn)
|
| 237 |
+
- (question_sd_o2_d2_fr -- answer_sd_o2_d2_fr)
|
| 238 |
+
- **O2+D3 Setting**:
|
| 239 |
+
- (question_td_o2_d3_yn -- answer_td_o2_d3_yn)
|
| 240 |
+
- (question_td_o2_d3_fr -- answer_td_o2_d3_fr)
|
| 241 |
+
- (question_sd_o2_d3_yn -- answer_sd_o2_d3_yn)
|
| 242 |
+
- (question_sd_o2_d3_fr -- answer_sd_o2_d3_fr)
|
| 243 |
+
- **Layout + O2 + D2 Setting**:
|
| 244 |
+
- (question_td_o2_d2_layout_yn -- answer_td_o2_d2_layout_yn)
|
| 245 |
+
- (question_td_o2_d2_layout_fr -- answer_td_o2_d2_layout_fr)
|
| 246 |
+
- (question_sd_o2_d2_layout_yn -- answer_sd_o2_d2_layout_yn)
|
| 247 |
+
- (question_sd_o2_d2_layout_fr -- answer_sd_o2_d2_layout_fr)
|
| 248 |
+
- **Layout + O2 + D3 Setting**:
|
| 249 |
+
- (question_td_o2_d3_layout_yn -- answer_td_o2_d3_layout_yn)
|
| 250 |
+
- (question_td_o2_d3_layout_fr -- answer_td_o2_d3_layout_fr)
|
| 251 |
+
- (question_sd_o2_d3_layout_yn -- answer_sd_o2_d3_layout_yn)
|
| 252 |
+
- (question_sd_o2_d3_layout_fr -- answer_sd_o2_d3_layout_fr)
|
| 253 |
|
| 254 |
+
### Question Types
|
| 255 |
|
| 256 |
+
- **'fr'**: Find Relation
|
| 257 |
+
- **'yn'**: Yes/No
|
| 258 |
|
| 259 |
+
### Points of View
|
| 260 |
|
| 261 |
+
- **'td'**: Top-Down View
|
| 262 |
+
- **'sd'**: North Facing View (from the south door)
|
| 263 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 264 |
|
| 265 |
---
|
| 266 |
## Relevant Paper
|