Welcome!

Description files and directories 📓

To reproduce article results you need to use this data:

• Detection and classification datasets
• Model weights for detection and classification 🚀 RECOMMENDED

test_cls.py - the file for testing classifier
train_cls.py - the file for training neural network

How to use train_cls.py.

Install all required libraries

$ pip install -r requrements_cls.txt

If you want to reproduce the results on "tigers_vs_leopards" dataset you need to specify path to config.yaml file that is located in configs/config.yaml. In this case, the weights that are trained on 29 classes will be used.

$ python train_cls.py -c configs/config.yaml

You can change some parameters like number of epochs, batch_size, input_size ant etc. The path to training and validation data is specified by default. If you want to use another dataset you should change train_dir, val_dir params.

$ python train_cls.py -c configs/config.yaml --epochs 10 --own_weights paht/to/your/weights --batch_size 32 --input_size 256 --loss smooth --train_dir path/to/your/train/data --val_dir path/to/your/val/data

When the network training is completed, directory results_train will be created, which will contain the following files:

• config.yaml - the configuration file with training params
• mapping.yaml - the file with classes on which the classifier was trained
• .pt - the weights of neural network
• results_.json - the results of training with losses and accuracy
• learning_curve.png - the curves of training and validation
• events.out.tfevents... - the results for tensorboard

How to use test_cls.py

If you want to evaluate the performance of a neural network on test data, you need to specify the path to the directory with the weights and test data

$ python test_cls.py --pt_w weights/Classification/tigers_vs_leopards/resnest101e --pt_data data/Classificationtigers_vs_leopards/test

or you can use your trained weights:

$ python test_cls.py --pt_w path/to/your/weights --pt_data data/Classification/tigers_vs_leopards/test

After testing, directory results_test will be created where the confusion matrix will be stored.

How to reproduce detection results

1. You need to clone git repository with YOLOv5

$ git clone https://github.com/ultralytics/yolov5.git

2. Then you need to install all libraries using requirements.txt for yolov5

$ pip install -r requirements.txt

3. Download dataset and change path to train and val dataset in animals.yaml file

4. After that, you need to train YOLOv5 using our pretrained weights

$ python train.py --imgsz 1280 --epochs 10 --data data/Detection/tigers_vs_leopards/animals.yaml --weights weights/Detection/YOLOv5_L6/weights/best.pt --single-cls --batch 2

4. To evaluate trained network on the test dataset you need to use val.py file in YOLOv5 repository

$ python val.py --test --imgsz 1280 --data data/Detection/tigers_vs_leopards/animals.yaml --weights weights/Detection/YOLOv5_L6/weights/best.pt --single-cls --batch 24 --single-cls