YOLOv8 models for deadwood segmentation from RGB UAV imagery

Model Details

Model Description

Model type: Instance segmentation
License: aGPL3
Finetuned from model: Ultralytics pretrained yolov8-seg models

Model Sources [optional]

Repository: https://github.com/mayrajeo/yolov8-deadwood
Paper: Added after submission
Demo: https://huggingface.co/spaces/mayrajeo/yolov8-deadwood

Uses

Direct Use

Models are meant for detecting and segmenting fallen and standing deadwood from RGB UAV images. As the models are trained on 640x640 pixel orthoimages with around 5 cm spatial resolution, they most likely work best with them.

Models can be directly used with ultralytics library like model = YOLO(<model_weights.pt>), and https://github.com/mayrajeo/yolov8-deadwood contains example scripts on how to use the models with larger orthomosaics.

Out-of-Scope Use

There are some things to keep in mind when using these models:

Models are trained using imagery from two geographically different locations, but both of the study sites consist of dense boreal forests in Finland.
The imagery was collected during leaf-on season, so the models will not produce optimal results during other seasons

How to Get Started with the Model

Single 640x640 pixel image chips can be processed with

from ultralytics import YOLO

model = YOLO(<path_to_model>)

res = model(<path_to_image>)

Larger orthomosaics should be processed with sahi library, or using the predict_image.py script from the related GitHub repository.

Training Details

Training Data

The models were trained on manually annotated deadwood polygon data. From Hiidenportti study area, 33 rectangular scenes were extracted and all visible deadwood was annotated from them. Same process was done fror Sudenpesänkangas, where 71 100x100 meter scenes were extracted.

In total, the dataset contained 13,813 deadwood instances, of which 2,502 were standing deadwood canopies and 11,311 were fallen deadwood trunks. Hiidenportti dataset contained 1,083 standing and 7,396 fallen annotations, whereas Sudenpesänkangas contained 1,419 standing and 3,915 fallen annotations.

As using the full sized scenes for training the models would be unfeasible due to their large sizes, the images were split into 640x640 pixel image chips without overlap, and the polygon annotations were converted to YOLO annotation format. After this process, the HP dataset contained 632 image chips for training, 142 for validating and 211 for testing, and SPK dataset contained 688, 224 and 224 chips for training, validating and testing respectively.

There are three types of models: models with _hp suffix are trained only on Hiidenportti data, models with _spk suffix only on Sudenpesänkangas data and models with _both suffix are trained on data from both sites.

Training Procedure

All models were trained on a single V100 GPGPU with 32GB of RAM on Puhti supercomputer hosted by CSC -- IT Center for Science, Finland.

Each model was trained for a maximum of 30 epochs with early stopping tolerance of 50 epohcs using Adam optimizer with initial learning rate of 0.001. Batch sizes for the models were chosen to be as large as possible so that they consumed a maximum of 60 % of the available GPU memory. Automatic mixed precision was used during training.

Evaluation

Testing Data, Factors & Metrics

Testing Data

Models were evaluated based on the test splits of both study sites.

Metrics

We used standard instance segmentation metrics, with the implementations from ultralytics library.

Results

Results for Hiidenportti test data

	precision(M) Total	precision(M) Fallen	precision(M) Standing	recall(M) Total	recall(M) Fallen	recall(M) Standing	mAP50(M) Total	mAP50(M) Fallen	mAP50(M) Standing	mAP50-95(M) Total	mAP50-95(M) Fallen	mAP50-95(M) Standing
yolov8n_hp	0.591	0.624	0.557	0.575	0.571	0.579	0.600	0.602	0.598	0.294	0.273	0.315
yolov8n_spk	0.512	0.560	0.463	0.469	0.485	0.454	0.464	0.495	0.433	0.198	0.194	0.202
yolov8n_both	0.720	0.741	0.699	0.571	0.534	0.607	0.647	0.612	0.683	0.317	0.263	0.371
yolov8s_hp	0.688	0.679	0.697	0.581	0.563	0.599	0.643	0.613	0.672	0.325	0.280	0.370
yolov8s_spk	0.548	0.669	0.428	0.478	0.463	0.492	0.484	0.528	0.439	0.212	0.213	0.211
yolov8s_both	0.650	0.623	0.678	0.614	0.644	0.584	0.656	0.638	0.675	0.324	0.284	0.364
yolov8m_hp	0.683	0.678	0.688	0.572	0.570	0.574	0.638	0.607	0.669	0.306	0.256	0.356
yolov8m_spk	0.609	0.702	0.516	0.563	0.539	0.587	0.551	0.591	0.512	0.256	0.254	0.258
yolov8m_both	0.676	0.643	0.710	0.619	0.637	0.602	0.671	0.638	0.703	0.338	0.286	0.390
yolov8l_hp	0.673	0.642	0.704	0.572	0.611	0.533	0.624	0.599	0.648	0.302	0.256	0.348
yolov8l_spk	0.609	0.700	0.518	0.530	0.524	0.536	0.544	0.585	0.504	0.254	0.254	0.253
yolov8l_both	0.701	0.658	0.744	0.622	0.627	0.616	0.676	0.648	0.705	0.339	0.291	0.386
yolov8x_hp	0.656	0.607	0.705	0.600	0.614	0.587	0.635	0.630	0.640	0.317	0.285	0.350
yolov8x_spk	0.550	0.706	0.395	0.493	0.460	0.526	0.469	0.548	0.390	0.211	0.234	0.188
yolov8x_both	0.709	0.684	0.734	0.620	0.603	0.638	0.682	0.654	0.709	0.353	0.306	0.400

Resuls for Sudenpesänkangas test data

	precision(M) Total	precision(M) Fallen	precision(M) Standing	recall(M) Total	recall(M) Fallen	recall(M) Standing	mAP50(M) Total	mAP50(M) Fallen	mAP50(M) Standing	mAP50-95(M) Total	mAP50-95(M) Fallen	mAP50-95(M) Standing
yolov8n_hp	0.683	0.492	0.873	0.233	0.249	0.218	0.308	0.288	0.329	0.138	0.106	0.170
yolov8n_spk	0.721	0.615	0.826	0.519	0.491	0.547	0.591	0.508	0.673	0.292	0.197	0.388
yolov8n_both	0.730	0.682	0.778	0.527	0.444	0.611	0.604	0.504	0.705	0.305	0.198	0.413
yolov8s_hp	0.586	0.446	0.726	0.342	0.347	0.336	0.414	0.331	0.497	0.187	0.121	0.253
yolov8s_spk	0.670	0.634	0.706	0.609	0.517	0.702	0.638	0.537	0.739	0.310	0.206	0.413
yolov8s_both	0.672	0.617	0.727	0.577	0.508	0.646	0.617	0.526	0.709	0.309	0.209	0.410
yolov8m_hp	0.613	0.440	0.786	0.339	0.330	0.349	0.407	0.331	0.482	0.185	0.122	0.248
yolov8m_spk	0.720	0.604	0.835	0.556	0.529	0.583	0.635	0.525	0.744	0.317	0.215	0.420
yolov8m_both	0.716	0.639	0.792	0.581	0.515	0.647	0.646	0.535	0.757	0.336	0.225	0.447
yolov8l_hp	0.573	0.414	0.732	0.340	0.366	0.313	0.397	0.328	0.465	0.162	0.113	0.212
yolov8l_spk	0.709	0.641	0.777	0.584	0.501	0.667	0.639	0.530	0.748	0.332	0.223	0.442
yolov8l_both	0.750	0.678	0.822	0.572	0.520	0.623	0.656	0.559	0.753	0.341	0.240	0.441
yolov8x_hp	0.675	0.543	0.807	0.322	0.362	0.282	0.421	0.385	0.457	0.185	0.141	0.229
yolov8x_spk	0.680	0.669	0.691	0.597	0.483	0.711	0.624	0.516	0.731	0.308	0.202	0.415
yolov8x_both	0.711	0.663	0.760	0.611	0.554	0.667	0.651	0.556	0.746	0.333	0.234	0.432

Citation [optional]

BibTeX:

Added after submitting

Model Card Contact

Janne Mäyrä, @mayrajeo on GitHub, Hugging Face and many other services.

mayrajeo
/

yolov8-deadwood