This repository holds the trained models and results for our modified YOLOv4 architectre. The training is done using the official DarkNet repository on VisDrone train portion of the dataset and training files are present in training_code folder. The trained weights for original and modified YOLOv4 architecture are available on Google Drive. The modified YOLOv4 architecture is shown in the below figure.
The testing is carried out using the official VisDrone DET Toolkit on VisDrone Test-Dev dataset. Detection files are generated using darknet_images.py file and resultant detection files are in the folders det_files_modified_yolov4 and det_files_original_yolov4. These detection files are passed into VisDrone DET Toolkit to get mAP results shown in the below figure.
