KuAvLab commited on
Commit
8e44a5c
·
verified ·
1 Parent(s): 7cfc14a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +94 -65
README.md CHANGED
@@ -1,88 +1,117 @@
1
  ---
2
  license: cc-by-nc-sa-4.0
3
  ---
 
 
 
 
 
 
4
  ## Introduction
5
- EMT is a comprehensive dataset for autonomous driving research, containing 57 minutes of diverse urban traffic footage from the Gulf Region. The dataset provides rich semantic annotations across two agent categories: people (pedestrians and cyclists), vehicles (seven classes). Each video segment spans 2.5-3 minutes, capturing challenging real-world scenarios:
 
 
 
6
 
7
- - **Dense Urban Traffic**: Complex multi-agent interactions in congested scenarios
8
- - **Weather Variations**: Clear and rainy conditions
9
- - **Visual Challenges**: High reflections from road surfaces and adverse weather combinations (rainy nights)
10
 
11
- The dataset provides dense annotations for:
12
- - **Detection & Tracking**: Multi-object tracking with consistent IDs - Available here
13
- - **Trajectory Prediction**: Future motion paths and social interactions - Refer to the github repo
14
- - **Intention Prediction**: Behavior understanding in complex scenarios - Refer to the github repo
15
 
 
 
16
 
17
- Validated through benchmarking on state-of-the-art models across tracking, trajectory prediction, and intention prediction tasks, with corresponding ground truth annotations for each benchmark.
18
 
 
19
 
20
- ## Quick start
21
- ``` bash
 
 
22
  from datasets import load_dataset
23
- # Load the dataset again with the 'force_redownload' option
24
- dataset = load_dataset("KuAvLab/EMT", split="train")#download_mode="reuse_cache_if_exists")# download_mode="force_redownload") or split train
25
- ```
26
- ### Avaiable labels:
27
- - Data from datatset has two outputs: image and object:
28
- - Image contains the frame image while object contains annotation:
29
- ``` # object labels
30
- bbox: bbox of detected objects
31
- track_id: tracking id of detected object
32
- class_id: class id of object
33
- class_name: type of object
34
  ```
35
 
36
- Sample use case:
 
 
 
 
 
 
 
 
 
 
 
 
37
  ```python
 
 
38
  for data in dataset:
39
  # Convert image from PIL to OpenCV format (BGR)
40
  img = np.array(data['image'])
 
41
  print("Classes:", data['objects']['class_name'])
42
- print("Bboxes:", len(data['objects']['bbox']),"\nTrack IDs:", data['objects']['track_id']),"\nclass IDs:", data['objects']['class_id']) )
 
 
43
  ```
44
 
 
45
 
46
- ### Data Collection
47
- | Aspect | Description |
48
- |:-------|:------------|
49
- | Duration | 57 minutes total footage |
50
- | Segments | 2.5-3 minutes continuous recordings |
51
- | FPS | 10fps for annotated frames |
52
- | Agent Classes | 2 Person classes and 7 Vehicle classes|
53
 
54
  ### Agent Categories
55
- 1. **People**
56
- - Pedestrians
57
- - Cyclists
58
-
59
- 2. **Vehicles**
60
- - Motorbike
61
- - Small motorised vehicle
62
- - Medium vehicle
63
- - Large vehicle
64
- - Car
65
- - Bus
66
- - Emergency vehicle
67
-
68
- ### Dataset Statistics
69
- | Category | Count |
70
- |----------|------------|
71
- | Annotated Frames | 34,386 |
72
- | Bounding Boxes | 626,634 |
73
- | Unique Agents | 9,094 |
74
- | Vehicle Instances | 7,857 |
75
- | Pedestrian Instances | 568 |
76
-
77
- | **Class** | **Description** | **Number of Bounding Boxes** | **Number of Agents** |
78
- |-----------|----------------|------------------------------|----------------------|
79
- | Pedestrian | An individual walking on foot. | 24,574 | 568 |
80
- | Cyclist | Any bicycle or electric bike rider. | 594 | 14 |
81
- | Motorbike | Includes motorcycles, bikes, and scooters with two or three wheels. | 11,294 | 159 |
82
- | Car | Any standard automobile. | 429,705 | 6,559 |
83
- | Small motorized vehicle | Motorized transport smaller than a car, such as mobility scooters and quad bikes. | 767 | 13 |
84
- | Medium vehicle | Includes vehicles larger than a standard car, such as vans or tractors. | 51,257 | 741 |
85
- | Large vehicle | Refers to vehicles larger than vans, such as lorries, typically with six or more wheels. | 37,757 | 579 |
86
- | Bus | Covers all types of buses, including school buses, single-deck, double-deck. | 19,244 | 200 |
87
- | Emergency vehicle | Emergency response units like ambulances, police cars, and fire trucks, distinguished by red and blue flashing lights. | 1,182 | 9 |
88
- | **_Overall:_** | | **576,374** | **8,842** |
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-nc-sa-4.0
3
  ---
4
+ ---
5
+ license: cc-by-nc-sa-4.0
6
+ ---
7
+
8
+ # EMT Dataset
9
+
10
  ## Introduction
11
+ EMT is a comprehensive dataset for autonomous driving research, containing **57 minutes** of diverse urban traffic footage from the **Gulf Region**. It includes rich semantic annotations across two agent categories:
12
+
13
+ - **People**: Pedestrians and cyclists
14
+ - **Vehicles**: Seven different classes
15
 
16
+ Each video segment spans **2.5-3 minutes**, capturing challenging real-world scenarios:
 
 
17
 
18
+ - **Dense Urban Traffic** – Multi-agent interactions in congested environments
19
+ - **Weather Variations** – Clear and rainy conditions
20
+ - **Visual Challenges** – High reflections and adverse weather combinations (e.g., rainy nights)
 
21
 
22
+ ### Dataset Annotations
23
+ This dataset provides annotations for:
24
 
25
+ - **Detection & Tracking** – Multi-object tracking with consistent IDs
26
 
27
+ For **intention prediction** and **trajectory prediction** annotations, please refer to our [GitHub repository](https://github.com/AV-Lab/emt-dataset).
28
 
29
+ ---
30
+
31
+ ## Quick Start
32
+ ```python
33
  from datasets import load_dataset
34
+
35
+ # Load the dataset
36
+ dataset = load_dataset("KuAvLab/EMT", split="train")
 
 
 
 
 
 
 
 
37
  ```
38
 
39
+ ### Available Labels
40
+ Each dataset sample contains two main components:
41
+
42
+ 1. **Image** – The frame image
43
+ 2. **Object** – The annotations for detected objects
44
+
45
+ #### Object Labels
46
+ - **bbox**: Bounding box coordinates (`x_min, y_min, x_max, y_max`)
47
+ - **track_id**: Tracking ID of detected objects
48
+ - **class_id**: Numeric class ID
49
+ - **class_name**: Object type (e.g., `car`, `pedestrian`)
50
+
51
+ #### Sample Usage
52
  ```python
53
+ import numpy as np
54
+
55
  for data in dataset:
56
  # Convert image from PIL to OpenCV format (BGR)
57
  img = np.array(data['image'])
58
+
59
  print("Classes:", data['objects']['class_name'])
60
+ print("Bboxes:", len(data['objects']['bbox']))
61
+ print("Track IDs:", data['objects']['track_id'])
62
+ print("Class IDs:", data['objects']['class_id'])
63
  ```
64
 
65
+ ---
66
 
67
+ ## Data Collection
68
+ | Aspect | Description |
69
+ |------------|----------------------------------|
70
+ | Duration | 57 minutes total footage |
71
+ | Segments | 2.5-3 minutes per recording |
72
+ | FPS | 10 fps for annotated frames |
73
+ | Agent Classes | 2 Person categories, 7 Vehicle categories |
74
 
75
  ### Agent Categories
76
+ #### **People**
77
+ - Pedestrians
78
+ - Cyclists
79
+
80
+ #### **Vehicles**
81
+ - Motorbike
82
+ - Small motorized vehicle
83
+ - Medium vehicle
84
+ - Large vehicle
85
+ - Car
86
+ - Bus
87
+ - Emergency vehicle
88
+
89
+ ---
90
+
91
+ ## Dataset Statistics
92
+ | Category | Count |
93
+ |-------------------|------------|
94
+ | Annotated Frames | 34,386 |
95
+ | Bounding Boxes | 626,634 |
96
+ | Unique Agents | 9,094 |
97
+ | Vehicle Instances | 7,857 |
98
+ | Pedestrian Instances | 568 |
99
+
100
+ ### Class Breakdown
101
+ | **Class** | **Description** | **Bounding Boxes** | **Unique Agents** |
102
+ |---------------------------|----------------|-------------------|----------------|
103
+ | Pedestrian | Walking individuals | 24,574 | 568 |
104
+ | Cyclist | Bicycle/e-bike riders | 594 | 14 |
105
+ | Motorbike | Motorcycles, bikes, scooters | 11,294 | 159 |
106
+ | Car | Standard automobiles | 429,705 | 6,559 |
107
+ | Small motorized vehicle | Mobility scooters, quad bikes | 767 | 13 |
108
+ | Medium vehicle | Vans, tractors | 51,257 | 741 |
109
+ | Large vehicle | Lorries, trucks (6+ wheels) | 37,757 | 579 |
110
+ | Bus | School buses, single/double-deckers | 19,244 | 200 |
111
+ | Emergency vehicle | Ambulances, police cars, fire trucks | 1,182 | 9 |
112
+ | **Overall** | | **576,374** | **8,842** |
113
+
114
+ ---
115
+
116
+ For more details , visit our [GitHub repository](https://github.com/AV-Lab/emt-dataset).
117
+