VisionTouch
- 2 Devlogs
- 20 Total hours
Incremental complexity use of ML algorithms and network models of gaze detection, eventually attached with hand coordination.
Finished the results for the model tests, and determined which models are best suited based on the lowest mean pixel error. The results of the head when still is very concerning because despite the more complex models coming to play, the ridge regression and linear regression still outperform complex models such as the decision tree, gradient boosting, and multilayer perceptrons.
Next course of action is to redo the datasets and introduce a larger number of points and maybe a dot ID to ensure that the dataset is split accurately and the amount is enough for accurate results.
Stage 1 and Stage 2 of VisionTouch focused on feature engineering and linear regression, and have concluded; the project started before hackatime was used. Ablation tests are shown for interest.
