Using YOLOv5 with W&B to Detect Starfish in the Great Barrier Reef

What we'll be looking for today: the crown-of-thorns-starfish

🧩 Introduction



📋 Table of Contents

📓 Code and notebooks

Train: Great-Barrier-Reef: YOLOv5 [train]
Inference: Great-Barrier-Reef: YOLOv5 [infer] 

🛠 Install libraries

!pip install -qU wandb

📌 One key point before we jump in

🪄 Initializing W&B



import wandb
wandb.login()

📝 Metadata

train_images/: The folder containing training set photos of the form video_{video_id}/{video_frame}.jpg
video_id: The ID number of the video the image was part of. (These video IDs are not meaningfully ordered.)
video_frame: The frame number of the image within the video. Note: you should expect to see occasional gaps in the frame number from when the diver surfaced.
sequence: The ID of a gap-free subset of a given video. The sequence ids are not meaningfully ordered.
sequence_frame: The frame number within a given sequence.
image_id: The ID code for the image, in the format {video_id}-{video_frame
annotations: The bounding boxes of any starfish detection in a string format that can be evaluated directly with Python. Does not use the same format as the predictions you will submit. Not available in test.csv. A bounding box is described by the pixel coordinate (x_min, y_min) of its lower left corner within the image together with its width and height in pixels --> (COCO format).

FOLD      = 1 # which fold to train
DIM       = 3000 
MODEL     = 'yolov5s'
BATCH     = 4
EPOCHS    = 15
OPTMIZER  = 'Adam'

PROJECT   = 'great-barrier-reef' # w&b in yolov5
NAME      = f'{MODEL}-dim{DIM}-fold{FOLD}' # w&b for yolov5

REMOVE_NOBBOX = True # remove images with no bbox
ROOT_DIR  = '/kaggle/input/tensorflow-great-barrier-reef/'
IMAGE_DIR = '/kaggle/images' # directory to save images
LABEL_DIR = '/kaggle/labels' # directory to save labels

Extracting our metadata

# Train Data
df = pd.read_csv(f'{ROOT_DIR}/train.csv')
df['old_image_path'] = f'{ROOT_DIR}/train_images/video_'+df.video_id.astype(str)+'/'+df.video_frame.astype(str)+'.jpg'
df['image_path']  = f'{IMAGE_DIR}/'+df.image_id+'.jpg'
df['label_path']  = f'{LABEL_DIR}/'+df.image_id+'.txt'
df['annotations'] = df['annotations'].progress_apply(eval)

df['num_bbox'] = df['annotations'].progress_apply(lambda x: len(x))

df['width']  = 1280
df['height'] = 720

🍚 Preparing our dataset

Creating labels for YOLOv5

One row per object
Each row is class `[x_center, y_center, width, height]` format.
Box coordinates must be in normalized `xywh` format (from `0 - 1`). If your boxes are in pixels, divide `x_center` and `width` by `image width`, and `y_center` and `height` by `image height
Class numbers are zero-indexed (start from `0`).

Configuration

The dataset root directory path and relative paths to `train / val / test` image directories (or *.txt files with image paths)
The number of classes nc and
A list of class names:['cots']Modify the yaml file below to update the hyper parameters.

names:
- cots
nc: 1
path: /kaggle/working
train: /kaggle/working/train.txt
val: /kaggle/working/val.txt

📦 YOLOv5 +🪄 W&B



!git clone https://github.com/awsaf49/yolov5-wandb.git yolov5 # clone
%cd yolov5
%pip install -qr requirements.txt  # install

import torch
import utils
display = utils.notebook_init()  # checks

🚅 Training

!python train.py --img {DIM}\
--batch {BATCH}\
--epochs {EPOCHS}\
--optimizer {OPTMIZER}\
--data /kaggle/working/gbr.yaml\
--weights {MODEL}.pt\
--project {PROJECT} --name {NAME} --entity ml-colabs\
--exist-ok

🔍 Results

Class Distribution

Batch Label/Prediction

Metrics

Loss

📊 Summary

Accurately identified starfish in real-time using YOLOv5 object detection model trained on underwater videos of coral reefs.
Demonstrated how to use YOLOv5 with W&B (Weights & Biases).
Analyzed the results and identified two key factors for improving performance: Large Image Size and Large Model Size.