search

d-j-kendall / beating_roulette

Designing a neural network to predict outcomes in roulette.
GNU Lesser General Public License v3.0
9 stars 2 forks source link

yolov3: Add additional parameter to specify transfer-learning starting weights #7

Closed d-j-kendall closed 5 years ago

d-j-kendall commented 5 years ago

Transfer learning always wants to start out at yolov3-spp.pt, but this may not always be the optimal case.

d-j-kendall commented 5 years ago

also --transfer learning switch does not work. The loss increases over time.

d-j-kendall commented 5 years ago

This was resolved in the most recent version of master, and transfer learning should work, but we do not need it as we are not training as large of a network now.