Open mez opened 5 years ago
You could start out by just visualizing the performance on entire sequences in the validation set. All you would need is y
and y_out
, both of size (N, 2)
where each row is [frame_no, track_id]
. The first array is the ground truth, and the second array is the model prediction.
Also, a track_id = -1
implies a false positive.
create the ability to view the graph as it is updating.