LettuceMOTS

Demo

demos.mp4

LettuceMOTS Dataset

The LettuceMOTS dataset is publicly available and can be found here.

Getting started

Prerequisites

Clone repo and create a Python>=3.7.0 environment, including PyTorch>=1.7. We recommend that you install Anaconda, set up a virtual environment and run:

git clone https://github.com/NanH5837/LettuceMOTS  # clone
cd LettuceMOTS
pip install -r requirements.txt  # install

Data

First, you need to download the LettuceMOTS dataset. For the convenience of training, we also provide the labels in the format required for YOLOv5 training. After the download is complete, place it in the following way:

LettuceMOTS
│   images
│   │    0000
│   │   │   000000.png
│   │   │   000001.png
│   │   │   ...
│   labels
│   │    0000
│   │   │   000000.txt
│   │   │   000001.txt
│   │   │   ...

After that, you need to create train.txt and val.txt. The content of the two files is the absolute path of the image in the above folder as follows:

/XXX/LettuceMOTS/images/0000/000000.png
/XXX/LettuceMOTS/images/0000/000001.png
...

Pretrained weights

We provide three pre-trained models trained on the LettuceMOTS dataset, and you can also choose a pre-trained model trained on COCO from YOLOv5 training. You can find three trained models here.

Train

After completing the above steps, you can perform the following operations to train the segmentation network:

cd segment
Python train.py --data lettuceMOTS.yaml --cfg yolov5m-seg.yaml  --batch-size 16 --weights yolov5m-seg.pt

Test

After training, you will get a best segmentation network model, then run the following to track：

cd tracking
Python track.py --weights best.pt --data lettuceMOTS.yaml --source /Path/to/your/souce

Reference

We borrow some code from YOLOv5 and SORT.