https://github.com/tronik27/roadsignsclassification
Implementation of a neural network for classifying a road sign image dataset
https://github.com/tronik27/roadsignsclassification
Last synced: 3 months ago
JSON representation
Implementation of a neural network for classifying a road sign image dataset
- Host: GitHub
- URL: https://github.com/tronik27/roadsignsclassification
- Owner: tronik27
- Created: 2021-07-30T12:40:07.000Z (almost 4 years ago)
- Default Branch: main
- Last Pushed: 2021-08-29T19:24:09.000Z (almost 4 years ago)
- Last Synced: 2024-12-27T23:42:45.014Z (5 months ago)
- Language: PureBasic
- Size: 22.4 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# RoadSignsClassification
This repository contains the implementation of the road sign classifier. The German Traffic Sign Recognition Benchmark dataset (https://benchmark.ini.rub.de) was used for training and testing. It contains about 39000 training and 12500 test images belonging to 43 classes. The complexity of this dataset lies in the rather strong imbalance of the classes (see the figure below) so the class weights are applied during training.
GTSRB dataset has this configuration:
```
├───Meta
├───Test
└───Train
├───0
├───1
...
├───41
└───42
```
When training a model, images for training are split into training and validation sets. Albumintations library (https://albumentations.ai/docs/api_reference/augmentations/) was used for image augmentation. As can be seen from the learning curves presented below, the classification quality metrics for the validation set were superior to those for the training set, which indicates that the model is not overfitting.
## Testing
This section presents the results of the trained model obtained on the test set:
*Loss: 0.13592;*
*Categorical accuracy: 98.779%;*
*F1 score: 98.780%;*
Classification report:
```
precision recall f1-score support
Speed limit (20km/h) 0.97 1.00 0.98 60
Speed limit (30km/h) 0.99 0.99 0.99 717
Speed limit (50km/h) 0.99 1.00 0.99 750
Speed limit (60km/h) 0.98 0.98 0.98 449
Speed limit (70km/h) 1.00 0.99 0.99 660
Speed limit (80km/h) 0.95 0.99 0.97 629
End of speed limit (80km/h) 1.00 0.82 0.90 150
Speed limit (100km/h) 1.00 0.99 0.99 450
Speed limit (120km/h) 0.99 0.99 0.99 449
No passing 0.98 1.00 0.99 479
No passing veh over 3.5 tons 1.00 1.00 1.00 659
Right-of-way at intersection 0.98 1.00 0.99 418
Priority road 1.00 0.99 0.99 688
Yield 1.00 1.00 1.00 719
Stop 1.00 1.00 1.00 270
No vehicles 0.96 0.99 0.97 210
Veh > 3.5 tons prohibited 0.99 1.00 1.00 150
No entry 1.00 1.00 1.00 360
General caution 1.00 0.95 0.97 389
Dangerous curve left 1.00 1.00 1.00 60
Dangerous curve right 0.95 1.00 0.97 90
Double curve 0.91 1.00 0.95 90
Bumpy road 1.00 1.00 1.00 120
Slippery road 0.99 0.99 0.99 150
Road narrows on the right 0.94 0.99 0.96 90
Road work 1.00 0.98 0.99 479
Traffic signals 0.97 1.00 0.99 180
Pedestrians 0.92 1.00 0.96 60
Children crossing 1.00 1.00 1.00 150
Bicycles crossing 1.00 1.00 1.00 90
Beware of ice/snow 0.98 0.89 0.93 150
Wild animals crossing 0.97 0.99 0.98 270
End speed + passing limits 1.00 1.00 1.00 60
Turn right ahead 1.00 1.00 1.00 209
Turn left ahead 0.99 1.00 1.00 119
Ahead only 1.00 1.00 1.00 389
Go straight or right 1.00 1.00 1.00 119
Go straight or left 0.98 1.00 0.99 60
Keep right 1.00 1.00 1.00 688
Keep left 0.98 0.98 0.98 90
Roundabout mandatory 1.00 0.94 0.97 90
End of no passing 0.81 0.85 0.83 60
End no passing veh > 3.5 tons 1.00 0.87 0.93 89
accuracy 0.99 12608
macro avg 0.98 0.98 0.98 12608
weighted avg 0.99 0.99 0.99 12608
```
Examples of image classification by model:
## Using script
If you want to train/evaluate or save the model by yourself, then use the code presented in **main.py**. The main parameters used in the code are specified in the file **config.py**. For image classification by pretrained model use **test.py**. The trained model is stored in the folder **trained_model/small_image_net10**. Below are examples of image classification by the trained model:
