Training Dataset and Evaluation Results of the Object Detector in "Purely vision-based collective movement of robots", Mezey et al.

The zip file contains the following data: 

TrainedModel folder: The trained “SSD MobileNet v2 FPNLite 320x320” model in tensorflow2 standard format (checkpoint, saved_model). Converted also as .tflite and fullinteger quantized for inference on edge TPU (saved_model/lite). For more information on the used model please see the README file within this folder.

ImageDataset folder: Labelled images used for training (train, valid) and evaluation (test). For further information see the README file in the folder (e.g. set size, download and augmentation/preprocessing). Images were labelled and data was augmented using Roboflow. Data is also available in the Roboflow model library, please see links in the README file within this folder.

EvaluationData folder: Data and images used for this evaluation report.

The evaluation results are summarized and interpreted in the README file in the root folder. 

This project was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy – EXC 2002/1 “Science of Intelligence” – project number 390523135.