File size: 2,075 Bytes
286fbb6 fdfb620 286fbb6 fdfb620 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 |
---
license: cc-by-4.0
tags:
- ocean
- benthic
- object-detection
---
# FathomNet2023 Baseline Model
## Model Details
- Trained by researchers at [Monterey Bay Aquarium Research Institute](https://www.mbari.org/) (MBARI) as a baseline for the [FathomNet2023 Competition](https://www.kaggle.com/competitions/fathomnet-out-of-sample-detection/overview) presented with the [Fine Grained Visual Categorization workshop](https://sites.google.com/view/fgvc10/home) at CVPR 2023.
- [Ultralytics YOLOv8.0.117](https://github.com/ultralytics/ultralytics/pull/3145)
- Object detection
- Fine tuned yolov8m to detect 290 fine grained taxonmic categories of benthic animals in the Greater Monterey Bay Area off the coast of Central California.
## Intended Use
- Make detections on images collect on the sea floor in the Monterey Bay Area.
## Factors
- Distribution shifts related to sampling platform, camera parameters, illumination, and deployment environment are expected to impact model performance.
- Evaluation was performed on an IID subset of available training data.
- Data to test out of distribution performance can be found on the [competition Kaggle page](https://www.kaggle.com/competitions/fathomnet-out-of-sample-detection/overview).
## Metrics
- [Precision-Recall curve](https://huggingface.co/FathomNet/MBARI-midwater-supercategory-detector/blob/main/plots/PR_curve.png) and [per class accuracy]((https://huggingface.co/FathomNet/MBARI-midwater-supercategory-detector/blob/main/plots/confusion_matrix.png)) were evaluated at test time.
- [email protected] = 0.33515
- Performance is quite variable depending on the target organism even when testing on in-distribution data.
## Training and Evaluation Data
- Training data is the [FathomNet2023 competition split](https://www.kaggle.com/competitions/fathomnet-out-of-sample-detection/overview) and internal MBARI data
- Class labels have a [long tail and localizations occur throughout the frame]().
## Deployment
In an environment running YOLOv8:
```
python classify/predict.py --weights best.pt --data data/images/
```
|