Datasets:
Modalities:
Image
Languages:
English
Size:
100M<n<1B
Tags:
computer-vision
autonomous-driving
driver-attention
gaze-estimation
semantic-segmentation
dataset
License:
Update README.md
Browse filesAdjustment of subheading spacing
README.md
CHANGED
|
@@ -53,13 +53,13 @@ Tester1/
|
|
| 53 |
│ └── ...
|
| 54 |
```
|
| 55 |
|
| 56 |
-
|
| 57 |
This directory contains **gaze–object annotation files** for each scene.
|
| 58 |
**File format**
|
| 59 |
- `<scene_id>`: scene index (starting from 0)
|
| 60 |
- Each row corresponds to one time step / frame
|
| 61 |
---
|
| 62 |
-
|
| 63 |
- `Stare_obj`: ID of the object being gazed at
|
| 64 |
- `0` indicates background or no valid gaze target
|
| 65 |
- `Stare_area`: coarse gaze region label on the screen
|
|
@@ -73,7 +73,7 @@ The screen resolution is **5740 × 1010**, and gaze regions are defined as:
|
|
| 73 |
| `MB` | Rear-view mirror | `[2890, 210] – [3540, 400]` |
|
| 74 |
|
| 75 |
---
|
| 76 |
-
|
| 77 |
- `Car{i}_screen_X`, `Car{i}_screen_Y`: 2D screen-space coordinates of risk-relevant vehicles
|
| 78 |
- Coordinates are aligned with the same frame as gaze annotations
|
| 79 |
- Missing vehicles are filled with `0`
|
|
@@ -87,14 +87,14 @@ These files jointly describe **where the driver is looking** and **where potenti
|
|
| 87 |
|
| 88 |
|
| 89 |
|
| 90 |
-
|
| 91 |
This directory contains **per-frame driving state and scene feature files** for each scene.
|
| 92 |
**File format**
|
| 93 |
- `<scene_id>`: scene index (starting from 0)
|
| 94 |
- Each row corresponds to one time step / frame
|
| 95 |
- Rows are time-aligned with gaze annotations and instance segmentation outputs
|
| 96 |
---
|
| 97 |
-
|
| 98 |
- `time`: timestamp (Unix time)
|
| 99 |
- `steering`: steering wheel angle
|
| 100 |
- `accelerator`: accelerator pedal value
|
|
@@ -103,11 +103,11 @@ This directory contains **per-frame driving state and scene feature files** for
|
|
| 103 |
- `Handchange_flag`: handover / control change indicator
|
| 104 |
- `Collision_flag`: collision indicator (binary)
|
| 105 |
---
|
| 106 |
-
|
| 107 |
- `main_car_id`: ID of the ego vehicle
|
| 108 |
- `main_car_x`, `main_car_y`: ego vehicle position in world coordinates
|
| 109 |
---
|
| 110 |
-
|
| 111 |
For each risk-relevant object in the scene, features are stored using indexed columns:
|
| 112 |
|
| 113 |
- Object indices: `Car1` … `Car9`
|
|
@@ -117,7 +117,7 @@ For each risk-relevant object in the scene, features are stored using indexed co
|
|
| 117 |
- Additional kinematic or geometric features
|
| 118 |
If a risk object is not present in a frame, its corresponding feature values are filled with `0`.
|
| 119 |
---
|
| 120 |
-
|
| 121 |
- `ScreenPoint2D_x`, `ScreenPoint2D_y`: projected 2D gaze point on the screen, aligned with gaze annotations
|
| 122 |
---
|
| 123 |
These files provide **low-level driving signals, ego vehicle states, and scene-level object features**, and are intended to be used jointly with:
|
|
|
|
| 53 |
│ └── ...
|
| 54 |
```
|
| 55 |
|
| 56 |
+
### `Gaze_object_output/`
|
| 57 |
This directory contains **gaze–object annotation files** for each scene.
|
| 58 |
**File format**
|
| 59 |
- `<scene_id>`: scene index (starting from 0)
|
| 60 |
- Each row corresponds to one time step / frame
|
| 61 |
---
|
| 62 |
+
##### Gaze Target Information
|
| 63 |
- `Stare_obj`: ID of the object being gazed at
|
| 64 |
- `0` indicates background or no valid gaze target
|
| 65 |
- `Stare_area`: coarse gaze region label on the screen
|
|
|
|
| 73 |
| `MB` | Rear-view mirror | `[2890, 210] – [3540, 400]` |
|
| 74 |
|
| 75 |
---
|
| 76 |
+
##### Vehicle Screen Positions
|
| 77 |
- `Car{i}_screen_X`, `Car{i}_screen_Y`: 2D screen-space coordinates of risk-relevant vehicles
|
| 78 |
- Coordinates are aligned with the same frame as gaze annotations
|
| 79 |
- Missing vehicles are filled with `0`
|
|
|
|
| 87 |
|
| 88 |
|
| 89 |
|
| 90 |
+
### `Tester1_feature_csv/`
|
| 91 |
This directory contains **per-frame driving state and scene feature files** for each scene.
|
| 92 |
**File format**
|
| 93 |
- `<scene_id>`: scene index (starting from 0)
|
| 94 |
- Each row corresponds to one time step / frame
|
| 95 |
- Rows are time-aligned with gaze annotations and instance segmentation outputs
|
| 96 |
---
|
| 97 |
+
##### Ego Vehicle and Driver State
|
| 98 |
- `time`: timestamp (Unix time)
|
| 99 |
- `steering`: steering wheel angle
|
| 100 |
- `accelerator`: accelerator pedal value
|
|
|
|
| 103 |
- `Handchange_flag`: handover / control change indicator
|
| 104 |
- `Collision_flag`: collision indicator (binary)
|
| 105 |
---
|
| 106 |
+
##### Ego Vehicle Position
|
| 107 |
- `main_car_id`: ID of the ego vehicle
|
| 108 |
- `main_car_x`, `main_car_y`: ego vehicle position in world coordinates
|
| 109 |
---
|
| 110 |
+
##### Surrounding Risk Object Features
|
| 111 |
For each risk-relevant object in the scene, features are stored using indexed columns:
|
| 112 |
|
| 113 |
- Object indices: `Car1` … `Car9`
|
|
|
|
| 117 |
- Additional kinematic or geometric features
|
| 118 |
If a risk object is not present in a frame, its corresponding feature values are filled with `0`.
|
| 119 |
---
|
| 120 |
+
##### Gaze Point Projection
|
| 121 |
- `ScreenPoint2D_x`, `ScreenPoint2D_y`: projected 2D gaze point on the screen, aligned with gaze annotations
|
| 122 |
---
|
| 123 |
These files provide **low-level driving signals, ego vehicle states, and scene-level object features**, and are intended to be used jointly with:
|