Project Purpose
ChartCV is a project aimed at automatically extracting data points from various chart types, including bar charts and scatter plots, using computer vision techniques.
Model Architecture
The model utilizes a convolutional neural network (CNN) architecture based on Resnet50fpn to detect and localize chart elements.
Data Size
The model was trained and tested on a dataset of 60k images, consisting of both synthetic and real-world charts.
Sample of image and elements plotted
Sample of image, prediction and target plotted
Key running hyperparameters include a learning rate of 0.005, a batch size of 8, and weight decay of 0.0005.
Future work
Future work will focus on extending the training with more epochs and implementing a learning rate scheduler and hyperparameter tuning. Additionally, we aim to expand the model's capabilities to handle more complex chart types and improve overall accuracy.