utils.py

import inspect
import re
from mako.template import Template
from mako import exceptions
from mako.lookup import TemplateLookup
import os
from scipy import misc
import numpy as np
import caffe
import matplotlib.pyplot as plt

# the filename for the root file that's being run.
def mfilename():
  stack = inspect.stack();
  filepath = (stack[1][1]);
  dotidx=[m.start() for m in re.finditer('\.', filepath)]
  filenm = filepath[0:dotidx[len(dotidx)-1]];
  dotidx=[m.start() for m in re.finditer('\/', filepath)]
  try:
    # this command fails in python <3
    filenm = filenm[dotidx[len(dotidx)-1]+1:];
  except:
    pass;
  return filenm;

def mkorender(tplfile,outfile,*args,**kwargs):
  mylookup = TemplateLookup(directories=[os.getcwd()])
  tpl=Template(filename=tplfile,lookup=mylookup);
  mkorendertpl(tpl,outfile,*args,**kwargs);

def mkorendertpl(tpl,outfile,*args,**kwargs):
  with open(outfile,"w") as f:
    try:
      f.write(tpl.render(*args,**kwargs));
    except:
      print((exceptions.text_error_template().render()))
      raise;

def mkorenderstr(strtpl,outfile,*args,**kwargs):
  mylookup = TemplateLookup(directories=[os.getcwd()])
  tpl=Template(strtpl,lookup=mylookup);
  mkorendertpl(tpl,outfile,*args,**kwargs);

def get_image(idx,dataset):
  im=misc.imread(dataset['dir'] + dataset['filename'][idx]);
  if (len(im.shape)==2):
    im=np.concatenate((im[:,:,None],im[:,:,None],im[:,:,None]),axis=2);
  return im;

def get_resized_image(idx,dataset,conf={}):
  targpixels=None;
  if 'targpixels' in dataset:
    targpixels=dataset['targpixels']
  if 'gri_targpixels' in conf:
    targpixels=conf['gri_targpixels'];
  maxdim=conf.get('gri_maxdim',None);
  im=get_image(idx,dataset);
  try:
    if(len(im.shape)==2):
      print("found grayscale image");
      im=np.array([im,im,im]).transpose(1,2,0);
    elif(im.shape[2]==4):
      print("found 4-channel png with channel 4 min "+str(np.min(im[:,:,3])));
      im=im[:,:,0:3];
  except:
    print("image id: " + str(idx));
    raise
    
  im=im.astype(np.float32)/255;
  if targpixels is not None:
    npixels = float(im.shape[0]*im.shape[1])
    # TODO: this has issues with the image boundary, as it samples zeros 
    # outside the image bounds.
    im=caffe.io.resize_image(im,(int(im.shape[0]*np.sqrt(targpixels/npixels)),int(im.shape[1]*np.sqrt(targpixels/npixels))))
  elif maxdim is not None:
    immax=max(im.shape[0:2]);
    ratio=float(maxdim)/float(immax);
    im=caffe.io.resize_image(im,(int(im.shape[0]*ratio),int(im.shape[1]*ratio)))
  return im;

def dispdata(net,idx,layer='data'):
  dat=net.blobs[layer].data[idx,:,:,:].transpose((1,2,0));
  dat=(dat-np.min(dat))/(np.max(dat)-np.min(dat));
  plt.imshow(dat);plt.show();