Computer vision
Quick-Draw Detection
A computer-vision experiment comparing CNN and Bi-LSTM approaches to classifying QuickDraw sketches. The CV reports approximately 92% test accuracy and a six-percentage-point improvement over a CNN-only baseline.
Approach
Compare model approaches
The project explores convolutional feature extraction alongside a recurrent model for sketch classification, using the QuickDraw dataset as its task domain.
Make the experiment reproducible
The CV reports the test result relative to a CNN-only baseline. A stronger technical write-up should also document the dataset split, preprocessing, class count and model configuration used for that result.
Outcome
A public ML project with a CV-reported test result of about 92% accuracy and a six-point improvement over the CNN-only baseline. Dataset and class-count details should be reconciled with the repository documentation.
- Python
- Computer vision
- CNN
- Bi-LSTM