lequytra / ssd Goto Github PK

• Cut down the number of layers used.

• Study the dimensions of our input (BatchSizes, 300,300,3) and the dimensions of our outputs.
• Create a function to generate random data: images, classes, bounding boxes , class + bounding boxes.

• Create a function to do this on a single scale.

• There are still some layers we have not added.
• Remove a few layers since the actual model is too big for us to train and evaluate.

The library openCV allows using webcam to do real-time predictions.

• There are existing API for parsing data for the SSD, using the COCO Dataset. We should write a code file that randomly sample data in specified batch sizes to feed in our model.

• Cut down the number of layers used.