Datasets:
Tasks:
Visual Question Answering
Formats:
json
Size:
< 1K
Tags:
video-question-answering
spatio-temporal-reasoning
furniture-assembly
multiple-choice
cvpr-2026
License:
Update questions/README.md
Browse files- questions/README.md +7 -4
questions/README.md
CHANGED
|
@@ -11,6 +11,9 @@ This directory contains the Flat-Pack Bench question set.
|
|
| 11 |
question with metadata for locating the corresponding furniture, video, prompt
|
| 12 |
media, and answer.
|
| 13 |
|
|
|
|
|
|
|
|
|
|
| 14 |
## `questions.jsonl` Schema
|
| 15 |
|
| 16 |
| Field | Type | Description |
|
|
@@ -23,11 +26,11 @@ media, and answer.
|
|
| 23 |
| `vid_category` | string | Furniture category, such as `Chair`, `Table`, `Bench`, `Shelf`, or `Misc`. |
|
| 24 |
| `furniture_name` | string | Furniture model name, such as `ronninge` or `stig`. |
|
| 25 |
| `video_id` | string | Video identifier used to align videos, frames, masks, and prompts. |
|
| 26 |
-
| `frame_idx` | integer or list | Prompt frame index for single-state questions, or a list for multi-state tracking questions. |
|
| 27 |
| `question` | object | Fully rendered multiple-choice question, options, and answer. |
|
| 28 |
-
| `prompt_img_fn` | string | Prompt image filename for non-tracking questions, when present. |
|
| 29 |
-
| `prompt_img_0_fn` | string | First prompt image filename for tracking/correspondence questions, when present. |
|
| 30 |
-
| `jumbled_prompt_img_1_fn` | string | Second prompt image filename for tracking/correspondence questions, when present. |
|
| 31 |
| `media_dir` | string | Optional generation/cache provenance path. Not required for interpreting the benchmark row. |
|
| 32 |
| `video` | string | Optional generated prompt/video filename from the benchmark pipeline. |
|
| 33 |
|
|
|
|
| 11 |
question with metadata for locating the corresponding furniture, video, prompt
|
| 12 |
media, and answer.
|
| 13 |
|
| 14 |
+
This file is used to construct the input text for the questions fed into the models for benchmarking, to ensure that every
|
| 15 |
+
model gets a consistent text input.
|
| 16 |
+
|
| 17 |
## `questions.jsonl` Schema
|
| 18 |
|
| 19 |
| Field | Type | Description |
|
|
|
|
| 26 |
| `vid_category` | string | Furniture category, such as `Chair`, `Table`, `Bench`, `Shelf`, or `Misc`. |
|
| 27 |
| `furniture_name` | string | Furniture model name, such as `ronninge` or `stig`. |
|
| 28 |
| `video_id` | string | Video identifier used to align videos, frames, masks, and prompts. |
|
| 29 |
+
| `frame_idx` | integer or list | Prompt frame index for single-state questions, or a list for multi-state tracking questions. Frame index is with respect to keyframe videos. |
|
| 30 |
| `question` | object | Fully rendered multiple-choice question, options, and answer. |
|
| 31 |
+
| `prompt_img_fn` | string | Prompt image filename for non-tracking questions, when present. Not required for interpreting the benchmark row. |
|
| 32 |
+
| `prompt_img_0_fn` | string | First prompt image filename for tracking/correspondence questions, when present. Not required for interpreting the benchmark row. |
|
| 33 |
+
| `jumbled_prompt_img_1_fn` | string | Second prompt image filename for tracking/correspondence questions, when present. Not required for interpreting the benchmark row. |
|
| 34 |
| `media_dir` | string | Optional generation/cache provenance path. Not required for interpreting the benchmark row. |
|
| 35 |
| `video` | string | Optional generated prompt/video filename from the benchmark pipeline. |
|
| 36 |
|