The dataset shared here is used to train the model for the team playing Goodgame, winning the Digital Race - Contest self-driving car programming for students in 2020. Each team’s task is to train and optimize AI models, and at the same time integrate them on a self-driving car model using NVIDIA’s Jetson TX2 board to follow. in the designated lanes, follow directional signs and avoid obstacles on the road. The Goodgame team, with a lot of talent and effort, won the championship. Goodgame captain Dat Vu helped share the source code and the data you used to train your AI models.
Digital Race 2020
1. Link to download data
The data is divided into two main tasks: Sign detection and lane detection.
2. Information about data
The data for sign detection is structured as follows:
Object Detection
Data
|βββ000000_10.png
|βββ000001_10.png
|βββ...
ββββ test.csv
train.csv
This dataset includes 12,764 images in the training set - training set and 2,561 images in the monitoring set - validation set, with 6 classes of signs:
- Turn left
- Turn right
- Go straight
- Stop
- No Turn Left
- No Turn Right
The labels of the dataset are contained in .csv
files with the following structure:
filename | xmin | ymin | xmax | ymax | class_id |
---|---|---|---|---|---|
00072.jpg | 148 | 53 | 159 | 63 | 4 |
Data for sign recognition
Lane Segment Dataset:
Directory structure:
Segmentation
GGDataSet
|βββ train_frames
|βββ train
|βββ train_000001.png
|βββ train_masks
|βββ train
|βββ train_000001.png
|βββ val_frames
|βββ val
|βββ val_000001.png
|βββ val_masks
|βββ val
|βββ val_000001.png
|βββ label_colors.txt
ββββ model_pb
models
train.py
ββββ convert_pb.py
The data set includes 6240 images in the training set - training set and 1448 images in the monitoring set - validation set, including segment labels for 3 layers: Background, Road lines and Road.
lane segment data
3. Source code
Those interested can see the notebooks for training deep learning networks using these datasets at the following link: https://github.com/makerhanoi/via-dataset. Goodgame also shares their source code to the community at here.