how to create a training dataset for image processing -
i have dataset of 10 jpeg high quality aerial images txt files containing information each vehicle's bounding box (width, height, angle, x & y axis,...). example:
@category:general @image:2012-04-26-muenchen-tunnel_4k0g0010.jpg #format: id type center.x center.y size.width size.height angle 0 30 1319 2338 35 11 56.451578 1 30 1337 2350 42 14 57.817368 2 30 224 3556 61 20 136.967797
how should create database of vehicles train in neural network using caffe? should use photoshop crop each vehicle , save them 1 one? or can use txt files create different classes of vehicles train in network sth matlab?
with many vehicles not hand. in python can load image numpy array , select boxes data provided in files. can handle angles rotating whole array , select box same way select 'normal' one.
if using different programming language should able follow approach need convert jpeg bitmap , somehow array.
i don't know caffe required capture exact bounderies because neural network needs boxes of same size.
Comments
Post a Comment