Questions on real-time python processing

[Invisible Arts]

I did

net.blobs['input'].data[...] =whatever
net.forward()
out = net.blobs['(last layer)']

and everything seemed to be working fine. But when I repeat that process on the same net the output is always the same with the former one. Is there a way to prevent this, or what's the right way (if I'm not doing things right) to conduct that process?

[Aske Rasch Lejbølle]

After running the network you need to copy the output to a new variable. What I do is:

net.blobs['data'].data[...] = whatever

out = net.forward() prob = out['prob'][...].copy() (or whatever layer you want output from)

Hope this helps.

[Jan C Peters]

Yes, I too suspect that you are using "out" again without copying the results first. Your last line out = net.blobs['(last layer)'] does not copy anything, it just creates a reference (or "pointer" if you will) called "out" to a blob datastructure. If the blob changes, out also changes, because it points only to the location of the data. Array data is only copied if requested explicitly (with .copy as suggested above) or when slicing the array.

Jan