TISC-Net: Towards the in-situ Trunk Identification and Length Measurement of Sea Cucumbers via Bezier Curve Modelling
- 🔭 I’m currently working on something cool.
Examples of trunk identification with the proposed TISC-Net. Outcomes of TISC-Net-m are presented. Each image is followed by the enlarged slice of the target, with B´ezier curve control points marked by solid circles. The images have been slightly resized for alignment. |
You need to install Ultralytics and configure your environment according to YOLOv8's official procedures.
Pip install the ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8.
pip install ultralytics| Name | Params(M) | download DUO | download ISTI |
|---|---|---|---|
| TISC-Net-m | 26.4 | DUO_model | ISTI_model |
| TISC-Net-xP6 | 99.3 | DUO_model | ISTI_model |
- You need to prepare datasets for following training and testing activities. Datasets
- The SC-ISTI dataset comprises 462 RGB images captured by an underwater robot in real habitat of sea cucumbers, with 670 sea cucumbers.
- The SC-DUO dataset consists of 1, 023 images containing 1, 856 sea cucumbers, which is created by selecting images containing sea cucumbers from DUO dataset.
- Pick a model and its config file, for example,
yolov8-pose-FEE.yaml. - Download the model m_DUO.pt
- Run the "test.py".
model = YOLO('/data/m_DUO.pt')
img_path = 'fig/2258.jpg'
- You need to replace the address of the model and images.
To train a model with "train.py", first prepare the custom dataset and set up the corresponding dataset yaml file sc_DUO-pose.yaml, then run "train.py":
from ultralytics import YOLO
# Load a model
model = YOLO('yolov8m-pose-FEE.yaml') # build a new model from YAML
# Train the model
model.train(data='sc_DUO-pose.yaml', epochs=100, imgsz=640, device = 3, batch = 32 )
metrics = model.val() # evaluate model performance on the validation set
metrics.box.map # map50-95
metrics.box.map50 # map50
metrics.box.map75 # map75
metrics.box.maps # a list con
You are my 