Help with pycaffe about dbpn-caffe

[alterzero]

Hi,
Sorry for my very late reply. Can you try this code?

##################
import numpy as np
from PIL import Image
import caffe
import sys
import scipy.io
import scipy.misc
import timeit

caffe.set_device(0)
caffe.set_mode_gpu()

inputsize=64
upscale=4

net = caffe.Net("model/DBPN_mat_4x.prototxt", "model/DBPN_4x.caffemodel",caffe.TEST)

transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2,0,1))
net.blobs['data'].reshape(1,3,inputsize,inputsize)

im = caffe.io.load_image('baby_GT_input.png', color=True)
net.blobs['data'].data[...] = transformer.preprocess('data', im)

start = timeit.default_timer()
out = net.forward()
stop = timeit.default_timer()
print stop - start

result=net.blobs['conv_final'].data

max_val, min_val = np.max(result[0]), np.min(result[0])

flatten_img = ((result.transpose((0,2,3,1)))[:] - min_val) / (max_val-min_val)

scipy.misc.imsave('result.png', flatten_img.reshape(1,1,inputsizeupscale,inputsizeupscale,3).swapaxes(1,2).reshape(inputsizeupscale,inputsizeupscale, 3))

[cmcuza]

Hi,

Thank you for your reply, it works perfect now !!!!!

[alexnie1997]

Hi,

I just copied the scripts you provided and did not change anything of them.

But I got an error as follows:

---

File "firsttrial.py", line 17, in
net.blobs['data'].data[...] = transformer.preprocess('data', im)
ValueError: could not broadcast input array from shape (3,57,86) into shape (1,3,64,64)

---

I don't know why you can run it without problem. Did you change some thing in the scripts?

Many thanks

[cmcuza]

Hi, you are right I had to change dimension in the prototxt, change 57 and 86 to 64.

[alexnie1997]

Thanks for your message.

After I changed these two parameters, I can run the code without errors.

Did you get the same output as that of pytorch version?

I found that the result of pycaffe is worse than that of pytorch.