In this project, we are aiming to implement [1] and improve it using Conv + LSTM layer for detection and tracking of the soccer ball for RoboCup Humanoid League.
example_data.csv contains information about training images with the following fields:
- image_file - image location,
- width - width of the image,
- height - height of the image,
- label - label of the object,
- xmin - top left x-coordinate of rectangle around object,
- ymin - top left y-coordinate of rectangle around object,
- xmax - bottom right x-coordinate of rectangle around object,
- ymax - bottom right y-coordinate of rectangle around object.
To extract and label images we are using Image Tagger We used YOLO for automatic ball detection and then we manually verified each the detection.
Note: One image file may contain multiple objects of different types.
- Train sweaty
python train.py --batch_size=16 --alpha=1000 --model_name=alpha1000 --epochs=50 - Test Sweaty
python test.py --load=pretrained_models/alpha1000_epoch_50.model --testSet=data/test/ --trainSet=data/train/
[1] Fabian Schnekenburger, Manuel Scharffenberg, Michael Wulker, Ulrich Hochberg, Klaus Dorer Detection and Localization of Features on a Soccer Field with Feedforward Fully Convolutional Neural Networks (FCNN) for the Adult-Size Humanoid Robot Sweaty