filter.py

import caffe
import cv2
#import matplotlib.image as mpimg
import numpy as np

caffe.set_mode_gpu()
net = caffe.Net('try.prototxt', 'mine.caffemodel', caffe.TEST)

image = cv2.imread('now5.jpg', 0) #gray
#print(image.shape)
np.set_printoptions(threshold=np.nan)


net.blobs['data'].reshape(1,1,356,356)
net.blobs['label'].reshape(1,24)
net.blobs['data'].data[...] = image 
net.forward()

#for layer_name, blob in net.blobs.items():
#	print(layer_name + '\t' + str(blob.data.shape))

#for layer_name, param in net.params.items():
#	print(layer_name + '\t' + str(param[0].data.shape), str(param[1].data.shape))

#print(net.blobs['conv1'].data.shape)
#print(net.blobs['conv1'].data)
#print(net.params['conv1'][0].data.shape)
filters = net.params['conv1'][0].data
#print(filters)

for i in range(32) :
	image = filters[i,0,:,:]
	minvalue = np.min(image)
	maxvalue = np.max(image)
	image=(image-minvalue)/(maxvalue-minvalue) * 255
	cv2.imwrite('image'+str(i)+'.jpg', image)
#for layer_name, blob in net.blobs.items():
#	print(layer_name + '\t' + str(blob.data.shape))
#print(image)
#label = net.blobs['label'].data
#output = net.blobs['ip2'].data

#print(label)
#print(output)