dataset_info:
features:
- name: file_name
dtype: image
- name: objects
dtype:
sequence:
- name: bbox
dtype:
sequence: int64
- name: category
dtype:
class_label:
names:
'0': surfer
'1': extra_class
configs:
- config_name: default
data_files:
- split: train
path:
- metadata_final_train.jsonl
- train/images/*
- split: valid
path:
- metadata_final_valid.jsonl
- valid/images/*
🌊 Wavez Pro: Real-Time Surfer Detection Dataset
📝 Project Explanation
This dataset is the core component of WAVEZ PRO, an innovative tech venture providing "Waze-like" real-time analytics for surfers. By utilizing computer vision, the model detects surfers in coastal environments to provide live crowd density data, helping the community find the best surf conditions.
📊 Exploratory Data Analysis (EDA)
The dataset consists of 8,194 high-quality images optimized for YOLO-based object detection.
1. Dataset Split
To ensure robust training and evaluation, the data is split into 80% training and 20% validation sets.
- Training Set: 6,590 images.
- Validation Set: 1,604 images.
2. Surfer Density & Distribution
Our statistical analysis (conducted on a large sample of the training data) reveals the complexity of the surf spot environments.
- Average Density: The dataset features an average of 8.71 surfers per image.
- Diversity: The data covers a spectrum from solitary surfers to crowded peak conditions with over 40 surfers per frame, providing the model with the necessary variety for real-world deployment.
🖼️ Visual Proof of Annotations
The following image represents a validation batch, demonstrating the high precision of the surfer class annotations (ID: 0).
Maintained by: Gilad Becher