FSOCO Dataset Sanitizer and YOLOv8 Trainer

This repository is your one-stop shop for preparing the FSOCO dataset, cleaning it up, augmenting it, and training a YOLOv8 model.

Let’s get started!

Instructions

1. Prerequisites (Before diving in, make sure you have):

   • Python 3.7 or higher
   • Virtual environment setup
   • A system that can handle YOLO training

2. Installation

   • Clone this repository Tip: Make sure you’re inside the repo directory before moving on.
   • Set up a virtual environment
   • Install the required libraries: should be in requirements.txt

3. Dataset Preparation

3.1 Download the FSOCO Dataset: Download the bounding boxes version from the FSOCO website.

• Extract the dataset and ensure the directory structure matches:
  • fsoco
    • train
      • annotations
      • images

Organized files = Happy training!

3.2 Convert COCO to YOLO Format: Run the coco_to_yolov.py script. This script takes the COCO-style annotations and converts them into YOLO-style .txt files.

3.3 Split into Train and Validation Sets: Run the split_train_val.py script to split the dataset into train and val directories.

• yolo_dataset
  • train
    • images
    • labels
  • val
    • images
    • labels

4. Data Augmentation The data augmentation process is currently ongoing.

5. Visualize the Dataset

   5.1 Install FiftyOne 5.2 Run the visualization script

6. YOLOv8 Training

   6.1 Create dataset.yaml

   • Place a dataset.yaml in the repository with the following structure: train: /path/to/yolo_dataset/train/images val: /path/to/yolo_dataset/val/images nc: <number_of_classes> names: [class1, class2, ...] 6.2 Start Training
     • Run the YOLOv8 training command

7. Results and Inference Results and inference functionalities are still pending implementation. Stay tuned for updates!