kitty7756 / cs7643-proj

CS7643-Project

Data Source

Dataset: https://console.cloud.google.com/storage/browser/quickdraw_dataset/full/simplified Format: ndjson Example Parser: https://github.com/googlecreativelab/quickdraw-dataset/blob/master/examples/nodejs/simplified-parser.js Useful Fields:

1. word: Category the player was prompted to draw.
2. drawing: A JSON array representing the vector drawing
3. recognized: Whether the word was recognized by the game.

Example ndjson: { "word":"airplane", "recognized":true, "drawing":[ [ [x0,x1,x2,x3,....], [y1,y1,y2,y3] ], [ [207,207,210,221,238], [74,103,114,128,135] ], [ [119,107,76,70,49,39,60,93], [72,41,3,0,1,5,38,70] ] ] }

We will pick 10 out of 345 labels and each with 100k+ dataset. We will randomly pick 80% as training set, 10% validation set, 10% test set.

Labels = [flower, apple, baseball, baskebtall, bird, book, bus, car, cat, dog]

How to Visualize dataset

https://stackoverflow.com/questions/63375078/visualize-google-quickdraw-data

RNN

To support batch processing, every dataset needs to have the same sequence length, so we can set a hard cutoff of 50 strokes. We will filter out data with more than 50 strokes. We will filter by recognized=true so that we can ensure the quality of the data is high.

CNN

We will filter by recognized=true so that we can ensure the quality of the data is high.