based on DeepWeeds (https://github.com/AlexOlsen/DeepWeeds)
This repository makes available the source code and public dataset for the work, "Weed Identification by Single-Stage and Two-Stage Neural Networks: A Study on the Impact of Image Resizers and Weights Optimization Algorithms", being published with open access by Frontiers in Plant Science: .
It contains annotated files for DeepWeeds dataset for various deep learning models using TensorFlow object detection API and YOLO/Darknet neural network framework. Also, the inference graph from the final/optimized DL model (Faster RCNN ResNet-101) is available.
It also contains configuration files for the deep learning models including SSD MobileNet, SSD Inception-v2, Faster RCNN ResNet-50, Faster RCNN ResNet-101, Faster RCNN Inception, Yolo-v4, RetinaNet, CenterNet ResNet-50, EfficientDet, and Yolo-v4.
The annotation files, inference graph, and source code are licensed under CC BY 4.0 license. The contents of this repository are released under an Apache 2 license.
The images were resized with annotations using resize-pascal-voc (https://github.com/italojs/resize_dataset_pascalvoc).
Due to the size of the images and models they are hosted outside of the Github repository.
- 300x300.rar (1.1 GB)
- 600x600.rar (2.8 GB)
- 640x640.rar (3.3 GB)
- 512x512.rar (2.0 GB)
- yolo-v4.rar (2.8 GB)
- Inference graph_final optimized model.rar (560 MB)
- inference_graph_enhanced_anchor_boxes.rar (560 MB)
This repository is a part of the PhD research of Muhammad Hammad Saleem (H.Saleem@massey.ac.nz; engr.hammadsaleem@gmail.com)
In case of any query, please contact Dr. Khalid Mahmood Arif (K.Arif@massey.ac.nz), Muhammad Hammad Saleem (H.Saleem@massey.ac.nz; engr.hammadsaleem@gmail.com)