StephanAkkerman/chart-info-yolo

Chart Info YOLO Dataset

This dataset contains annotated screenshots of financial charts (e.g. TradingView), formatted for object detection.
It’s designed to train small YOLO models that detect UI elements used for downstream OCR:

• 0 = symbol_title — the title block with the ticker and name at the top of the chart
• 1 = last_price_pill — the rounded price pill on the right-side price axis (current/last price)

Structure

The dataset is provided in the classic YOLO structure:

images/
  train/*.png
  val/*.png
  test/*.png
labels/
  train/*.txt
  val/*.txt
  test/*.txt
data.yml

Images

In the images directory you will find the chart images which is a sample from https://huggingface.co/datasets/StephanAkkerman/stock-charts.

Labels

The labels directory provides the bounding boxes for the symbol title and last price pill for each chart.

<class_id> <x_center> <y_center> <width> <height>

All coordinates are normalized to [0, 1]. Some charts are unlabeled as I have only focused on TradingView charts for the labeling.

Example

The following image is an example of what a labeled chart looks like.

Usage

Use the following example to download the dataset to use for YOLO model training.

from huggingface_hub import snapshot_download

snapshot_download(
  repo_id="StephanAkkerman/chart-info-yolo",
  repo_type="dataset",
  local_dir="datasets/tradingview",
  local_dir_use_symlinks=False,
)

After this you need to refer to the data.yml path for the training and then it works.

yolo detect train model=yolo12n.pt data=datasets/tradingview/data.yaml imgsz=1792 epochs=80

Intended use

This dataset is built to support:

• detecting chart UI elements (symbol_title, last_price_pill)
• cropping them for OCR (e.g. PaddleOCR) to extract ticker, name, and current price

Contributions (extra chart sources, more UI element classes) are welcome.