notebook.community

Edit and run

Licensed under the Apache License, Version 2.0 (the "License");





In [0]:



    


# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

Decomposed Feature Visualization (rough)

Decomposed feature visualization is a techhnique for understanding the mechanics of how neural network features operate. This is a very early experiment with them.

Install, Import, Load Model





In [0]:



    


# Install Lucid

!pip install --quiet lucid==0.0.5
#!pip install --quiet --upgrade-strategy=only-if-needed git+https://github.com/tensorflow/lucid.git



    











In [0]:



    


# Imports

import numpy as np
import tensorflow as tf

import lucid.modelzoo.vision_models as models
from lucid.misc.io import show
import lucid.optvis.objectives as objectives
import lucid.optvis.param as param
import lucid.optvis.render as render
import lucid.optvis.transform as transform



    











In [0]:



    


import scipy.ndimage as nd



    











In [0]:



    


# Let's import a model from the Lucid modelzoo!

model = models.InceptionV1()
model.load_graphdef()



    











In [0]:



    


def hidden_opt(model, layer, obj, W, require_pos=False):
  with tf.Graph().as_default(), tf.Session() as sess:
    
    t_input = tf.placeholder("float32", [1, None, None, 3])
    T = render.import_model(model, t_input, t_input)
    loss = obj(T)
    t_layer = T(layer)
    vecs = 0.1*np.random.randn(1, W, W, t_layer.shape[-1])
    t_layer_grad = tf.gradients(loss, t_layer)[0]

    for i in range(1000):
      if i % 250 == 0: print loss.eval({t_layer: vecs}), 
      vecs += 0.1 * t_layer_grad.eval({t_layer: vecs})
      if require_pos:
        vecs = np.maximum(0, vecs)
    print ""
      
    return vecs



    











In [0]:



    


def decomposed_vis(layer, obj, W=8, W2=32):
  vecs = hidden_opt(model, layer, obj, W=W)
  vecs_flat = vecs.reshape([-1, vecs.shape[-1]])
  
  mags = np.sqrt((vecs[0]**2).sum(2))

  param_f = lambda: param.image(W2, batch=W*W)
  obj = objectives.Objective.sum([
      objectives.direction(layer, vec, batch=n)
      for n, vec in enumerate(vecs_flat)
  ])

  imgs = render.render_vis(model, obj, param_f, thresholds=(512,))
  
  imgs_ = np.pad(imgs[0].reshape([W,W,W2,W2,3]), [(0,0), (0,0), (0,1), (0,1), (0,0)], mode='constant', constant_values=(1,))
  img_grid = np.hstack(np.hstack(imgs_))
  mags_grid = np.hstack(np.hstack(
      np.pad(
          nd.zoom(mags.reshape([W,W,1,1])/mags.max(), [1, 1, W2, W2], order=0)[..., None]
          , [(0,0), (0,0), (0,1), (0,1), (0,0)], mode='constant', constant_values=(1,))

  ))
  #show(mags_grid)
  #show(img_grid)
  show(img_grid * mags_grid  + 0.8*(1-mags_grid) )



    











In [8]:



    


for node in model.graph_def.node:
  if not "/" in node.name and not "_" in node.name:
    print node.name



    


















    






input
conv2d0
maxpool0
localresponsenorm0
conv2d1
conv2d2
localresponsenorm1
maxpool1
mixed3a
mixed3b
maxpool4
mixed4a
mixed4b
mixed4c
mixed4d
mixed4e
maxpool10
mixed5a
mixed5b
avgpool0
nn0
softmax0
nn1
softmax1
softmax2
output
output1
output2

















In [7]:



    


obj = objectives.neuron("mixed4c_pre_relu", 447)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4b", obj, W=5, W2=48)



    


















    






512 [5920.5444]










    
















    






[-0.5783826] [27.253613] [55.562912] [83.730484] 
512 712.68896










    
















    























In [9]:



    


obj = objectives.neuron("mixed4c_pre_relu", 447)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=8, W2=48)



    


















    






512 [5289.624]










    
















    






[-2.8396463] [47.62098] [82.608086] [117.355286] 
512 1397.7031










    
















    























In [10]:



    


obj = objectives.neuron("mixed4c_pre_relu", 447)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed3a", obj, W=14, W2=32)



    


















    






512 [5088.3745]










    
















    






[-5.9211655] [41.485485] [93.910065] [175.572] 
512 1621.6318










    
















    























In [47]:



    


|fobj = objectives.neuron("mixed4d_pre_relu", 426)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4c", obj, W=5, W2=48)



    


















    






512 [4948.5024]










    
















    






[0.07394248] [39.683586] [77.32095] [114.89942] 
512 651.6599










    
















    























In [53]:



    


obj = objectives.neuron("mixed4b_pre_relu", 368)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=48)



    


















    






512 [3244.923]










    
















    






[0.37219653] [14.156713] [25.67627] [37.20147] 
512 371.83102










    
















    























In [54]:



    


obj = objectives.neuron("mixed4b_pre_relu", 369)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=48)



    


















    






512 [2491.0662]










    
















    






[0.22406384] [17.83633] [32.860977] [47.850792] 
512 384.26587










    
















    























In [9]:



    


obj = objectives.neuron("mixed4b_pre_relu", 396)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=70)



    


















    






512 [3838.2075]










    
















    






[0.32719594] [13.5257845] [24.346987] [35.177425] 
512 500.67758










    
















    























In [10]:



    


obj = objectives.neuron("mixed4b_pre_relu", 402)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=70)



    


















    






512 [2667.6453]










    
















    






[0.49565056] [12.310391] [22.122395] [31.906187] 
512 409.88477










    
















    























In [11]:



    


obj = objectives.neuron("mixed4b_pre_relu", 403)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=70)



    


















    






512 [4483.091]










    
















    






[0.20701264] [19.705093] [36.81204] [53.980072] 
512 749.6432










    
















    























In [12]:



    


obj = objectives.neuron("mixed4b_pre_relu", 405)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=70)



    


















    






512 [4164.884]










    
















    






[-0.24503744] [15.056617] [28.254328] [41.40747] 
512 630.4484










    
















    























In [21]:



    


obj = objectives.neuron("mixed4b_pre_relu", 409)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [4871.179]










    
















    






[0.17435488] [18.935804] [35.22211] [51.481693] 
512 821.6135










    
















    























In [20]:



    


obj = objectives.neuron("mixed4b_pre_relu", 414)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [3360.9058]










    
















    






[-0.01797339] [11.983153] [21.800112] [31.748919] 
512 509.64902










    
















    























In [19]:



    


obj = objectives.neuron("mixed4b_pre_relu", 418)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [4555.3804]










    
















    






[-0.15576442] [28.00195] [53.289303] [78.57025] 
512 942.2225










    
















    























In [22]:



    


obj = objectives.neuron("mixed4b_pre_relu", 418)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [5914.3447]










    
















    






[-0.0299177] [27.79535] [53.048737] [78.327034] 
512 930.49243










    
















    























In [23]:



    


obj = objectives.neuron("mixed4b_pre_relu", 429)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [5199.784]










    
















    






[0.0800992] [19.527641] [36.2039] [52.909515] 
512 755.61194










    
















    























In [24]:



    


obj = objectives.neuron("mixed4b_pre_relu", 437)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [3142.1343]










    
















    






[0.28221816] [12.072634] [22.223206] [32.332207] 
512 597.3205










    
















    























In [7]:



    


obj = objectives.neuron("mixed4b_pre_relu", 450)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=50)



    


















    






512 [3270.2854]










    
















    






[0.06421056] [15.12474] [29.023287] [42.880196] 
512 322.58386










    
















    























In [8]:



    


obj = objectives.neuron("mixed4c_pre_relu", 409)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4b", obj, W=5, W2=70)



    


















    






512 [3405.1697]










    
















    






[-0.7962754] [23.149366] [44.361168] [66.311935] 
512 514.877










    
















    























In [9]:



    


obj = objectives.neuron("mixed4c_pre_relu", 418)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4b", obj, W=5, W2=70)



    


















    






512 [3191.9436]










    
















    






[-0.22031027] [14.297702] [26.172873] [38.086796] 
512 391.03894










    
















    























In [18]:



    


obj = objectives.neuron("mixed4b_pre_relu", 462)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=70)



    


















    






512 [2753.4014]










    
















    






[-0.05936106] [12.628906] [24.137106] [35.597] 
512 319.17712










    
















    























In [57]:



    


obj = objectives.neuron("mixed4b_pre_relu", 373)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=70)



    


















    






512 [2746.9575]










    
















    






[-0.24139129] [18.711418] [34.711914] [50.716488] 
512 418.32898










    
















    























In [55]:



    


obj = objectives.neuron("mixed4b_pre_relu", 373)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=5, W2=48)



    


















    






512 [3042.0288]










    
















    






[-0.06549181] [18.982788] [35.06047] [51.15104] 
512 456.83997










    
















    























In [52]:



    


obj = objectives.neuron("mixed4d_pre_relu", 426)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("mixed4a", obj, W=10, W2=48)



    


















    






512 [5159.9834]










    
















    






 [-4.6794705] [52.082783] [88.43633] [124.26673] 
512 1134.271










    
















    























In [50]:



    


obj = objectives.neuron("mixed4a_pre_relu", 368)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=38)



    


















    






512 [4599.4194]










    
















    






[-0.34192824] [27.047497] [51.625256] [76.255486] 
512 814.5455










    
















    























In [48]:



    


obj = objectives.neuron("mixed4a_pre_relu", 368)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [5427.123]










    
















    






[-0.48527247] [27.11984] [51.837227] [76.59127] 
512 726.41345










    
















    























In [11]:



    


obj = objectives.neuron("mixed4a_pre_relu", 460)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [2297.1968]










    
















    






[-0.0648718] [19.397154] [37.41374] [55.415348] 
512 624.3943










    
















    























In [10]:



    


obj = objectives.neuron("mixed4a_pre_relu", 475)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [1931.3134]










    
















    






[0.23064563] [23.879513] [45.882442] [67.87124] 
512 555.57635










    
















    























In [8]:



    


obj = objectives.neuron("mixed4a_pre_relu", 455)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [1745.6112]










    
















    






[0.17473549] [20.072954] [38.778442] [57.439262] 
512 314.85904










    
















    























In [7]:



    


obj = objectives.neuron("mixed4a_pre_relu", 443)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [2629.794]










    
















    






[0.21602516] [11.583387] [21.114212] [30.644304] 
512 529.1863










    
















    























In [14]:



    


obj = objectives.neuron("mixed4a_pre_relu", 426)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3772.6838]










    
















    






[-0.1787216] [9.528883] [16.943613] [24.247198] 
512 905.1608










    
















    























In [42]:



    


obj = objectives.neuron("mixed4a_pre_relu", 371)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3121.1353]










    
















    






[0.00719921] [25.367666] [48.19033] [71.033295] 
512 503.2522










    
















    























In [49]:



    


obj = objectives.neuron("mixed4a_pre_relu", 372)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3168.1628]










    
















    






[0.57510674] [24.82596] [46.552643] [68.34562] 
512 507.82306










    
















    























In [35]:



    


obj = objectives.neuron("mixed4a_pre_relu", 374)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3070.728]










    
















    






[-0.06418443] [28.307547] [53.46128] [78.73724] 
512 501.6889










    
















    























In [43]:



    


obj = objectives.neuron("mixed4a_pre_relu", 375)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3330.3484]










    
















    






[-0.00951466] [17.531813] [32.74345] [48.010677] 
512 467.01202










    
















    























In [36]:



    


obj = objectives.neuron("mixed4a_pre_relu", 383)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3866.8867]










    
















    






[-0.5437354] [30.695196] [58.576145] [86.52927] 
512 645.2164










    
















    























In [37]:



    


obj = objectives.neuron("mixed4a_pre_relu", 386)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3329.6033]










    
















    






[-0.03198454] [22.326384] [41.84923] [61.446278] 
512 514.9338










    
















    























In [44]:



    


obj = objectives.neuron("mixed4a_pre_relu", 389)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=38)



    


















    






512 [2488.431]










    
















    






[0.07138123] [21.141193] [39.561634] [57.986786] 
512 443.3991










    
















    























In [38]:



    


obj = objectives.neuron("mixed4a_pre_relu", 389)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [2607.3064]










    
















    






[-0.28256202] [20.985905] [39.390106] [57.799294] 
512 478.9345










    
















    























In [39]:



    


obj = objectives.neuron("mixed4a_pre_relu", 391)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=38)



    


















    






512 [6638.676]










    
















    






[-0.63946664] [74.706894] [145.8053] [216.96138] 
512 737.2148










    
















    























In [40]:



    


obj = objectives.neuron("mixed4a_pre_relu", 391)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [5348.0854]










    
















    






[-0.74396044] [74.49679] [145.59766] [216.71626] 
512 792.11835










    
















    























In [59]:



    


obj = objectives.neuron("mixed4a_pre_relu", 397)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3383.6567]










    
















    






[0.32338637] [18.938078] [35.517536] [52.08198] 
512 681.6991










    
















    























In [18]:



    


obj = objectives.neuron("mixed4a_pre_relu", 409)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3189.899]










    
















    






[0.03085956] [13.365607] [23.686354] [33.939785] 
512 1194.0269










    
















    























In [22]:



    


obj = objectives.neuron("mixed4a_pre_relu", 408)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=38)



    


















    






512 [2473.8054]










    
















    






[0.68488955] [12.153672] [21.703093] [31.147547] 
512 981.6162










    
















    























In [19]:



    


obj = objectives.neuron("mixed4a_pre_relu", 408)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [2658.7944]










    
















    






[0.7928688] [12.229861] [21.781069] [31.226694] 
512 1125.5286










    
















    























In [45]:



    


obj = objectives.neuron("mixed4a_pre_relu", 402)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [2728.9512]










    
















    






[0.06034447] [14.722796] [27.360315] [39.986626] 
512 693.60034










    
















    























In [46]:



    


obj = objectives.neuron("mixed4a_pre_relu", 423)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [3321.3325]










    
















    






[-0.02896704] [19.680662] [37.039642] [54.39272] 
512 798.3628










    
















    























In [17]:



    


obj = objectives.neuron("mixed4a_pre_relu", 417)
render.render_vis(model, obj, lambda: param.image(100))
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






512 [2232.0295]










    
















    






[0.01840711] [13.035285] [23.652243] [34.135235] 
512 1019.05347










    
















    























In [9]:



    


_ = render.render_vis(model, "mixed4a_pre_relu:476")



    


















    






512 1288.9688










    























In [41]:



    


obj = objectives.neuron("mixed3a_pre_relu", 188)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("maxpool1", obj, W=5, W2=32)



    


















    






512 [672.3662]










    
















    






[-0.11067072] [86.355125] [170.51508] [254.75972] 
512 435.52533










    
















    























In [34]:



    


obj = objectives.neuron("mixed3a_pre_relu", 188)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("maxpool1", obj, W=5, W2=32)



    


















    






512 [579.0347]










    
















    






[-0.08502422] [51.642185] [99.93456] [148.23859] 
512 571.719










    
















    























In [19]:



    


obj = objectives.neuron("mixed3b_pre_relu", 318)
decomposed_vis("mixed3a", obj, W=5, W2=48)



    


















    






[-0.4931006] [18.95018] [34.11355] [49.26719] 
512 674.20184










    
















    
















    
















    























In [58]:



    


obj = objectives.neuron("mixed3b_pre_relu", 342)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("mixed3a", obj, W=5, W2=32)



    


















    






512 [1682.0072]










    
















    






[-0.76788425] [49.55262] [95.37189] [141.20833] 
512 639.98145










    
















    























In [17]:



    


obj = objectives.neuron("mixed3b_pre_relu", 316)
decomposed_vis("mixed3a", obj, W=5, W2=48)



    


















    






[0.18203844] [25.392897] [47.812107] [70.320496] 
512 1006.4719










    
















    
















    
















    























In [15]:



    


obj = objectives.neuron("mixed3b_pre_relu", 320)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("mixed3a", obj, W=5, W2=32)



    


















    






512 [1267.8467]










    
















    






[-0.23418218] [19.550758] [35.001556] [50.362965] 
512 904.236










    
















    
















    
















    























In [14]:



    


obj = objectives.neuron("mixed3b_pre_relu", 339)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("mixed3a", obj, W=5, W2=32)



    


















    






512 [2043.5299]










    
















    






[-0.5394805] [23.07312] [42.313976] [61.448772] 
512 1018.1162










    
















    
















    
















    























In [8]:



    


obj = objectives.neuron("mixed3b_pre_relu", 367)
decomposed_vis("mixed3a", obj, W=5, W2=48)



    


















    






[-0.4086998] [25.034378] [45.71674] [66.32295] 
512 1025.3914










    
















    
















    
















    























In [17]:



    


obj = objectives.neuron("mixed3b_pre_relu", 344)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("mixed3a", obj, W=5, W2=32)



    


















    






512 [1465.3738]










    
















    






[-0.35458434] [19.131765] [34.565754] [49.921234] 
512 936.8495










    
















    
















    
















    























In [23]:



    


obj = objectives.neuron("mixed3a_pre_relu", 174)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("maxpool1", obj, W=5, W2=32)



    


















    






512 [444.72144]










    
















    






[-0.48838598] [60.01479] [116.77704] [173.53343] 
512 590.5175










    
















    























In [12]:



    


obj = objectives.neuron("mixed3a_pre_relu", 149)
decomposed_vis("maxpool1", obj, W=5, W2=48)



    


















    






[-0.31444335] [63.38946] [123.50652] [183.62169] 
512 672.0368










    
















    
















    
















    























In [11]:



    


obj = objectives.neuron("mixed3a_pre_relu", 177)
decomposed_vis("maxpool1", obj, W=5, W2=48)



    


















    






[-0.48851967] [46.67312] [90.73688] [134.84222] 
512 608.87915










    
















    
















    
















    























In [10]:



    


obj = objectives.neuron("mixed4b_pre_relu", 175)
decomposed_vis("mixed4a", obj, W=5, W2=48)



    


















    






[-0.32742932] [20.790764] [38.629528] [56.37598] 
512 710.4025










    
















    
















    
















    























In [18]:



    


obj = objectives.neuron("mixed4b_pre_relu", 176)
render.render_vis(model, obj, lambda: param.image(80))
decomposed_vis("mixed4a", obj, W=3, W2=42)



    


















    






512 [3725.3474]










    
















    






[0.07476339] [14.936309] [27.064476] [39.07512] 
512 855.62634










    
















    
















    
















    























In [16]:



    


obj = objectives.neuron("mixed4b_pre_relu", 235)
decomposed_vis("mixed4a", obj, W=5, W2=32)



    


















    






[0.1318711] [13.673563] [24.048182] [34.356586] 
512 278.79385










    
















    
















    
















    























In [19]:



    


obj = objectives.neuron("mixed4b_pre_relu", 237)
decomposed_vis("mixed4a", obj, W=5, W2=32)



    


















    






[-0.02376008] [15.242497] [27.379873] [39.429676] 
512 261.1907










    
















    
















    
















    























In [17]:



    


obj = objectives.neuron("mixed4c_pre_relu", 8)
decomposed_vis("mixed4b", obj, W=3, W2=32)



    


















    






[0.7051698] [16.485058] [31.993143] [47.500263] 
512 138.37042










    
















    
















    
















    























In [18]:



    


obj = objectives.neuron("mixed4c_pre_relu", 8)
decomposed_vis("mixed4a", obj, W=5, W2=32)



    


















    






[0.8526461] [15.270745] [28.0152] [40.75976] 
512 231.13649










    
















    
















    
















    























In [26]:



    


obj = objectives.neuron("mixed4a_pre_relu", 36)
decomposed_vis("maxpool4", obj, W=5, W2=32)



    


















    






[0.38001198] [18.185] [35.826694] [53.467373] 
512 431.58755










    
















    
















    
















    























In [12]:



    


obj = objectives.neuron("mixed4a_pre_relu", 36)
decomposed_vis("mixed3a", obj, W=10, W2=18)



    


















    






[0.02971543] [8.785434] [24.014843] [40.957222] 
512 674.4531










    
















    
















    
















    























In [13]:



    


obj = objectives.neuron("mixed4a_pre_relu", 36)
decomposed_vis("mixed3b", obj, W=10, W2=18)



    


















    






[0.20501895] [19.109838] [36.76633] [54.410286] 
512 139.7491










    
















    
















    
















    























In [8]:



    


obj = objectives.channel("mixed4a_pre_relu", 36)
decomposed_vis("mixed3a", obj, W=10, W2=18)



    


















    






0.07912122 0.8464691 2.0357249 3.9925823 
512 373.9257










    
















    
















    
















    























In [9]:



    


obj = objectives.channel("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=16)



    


















    






0.1286154 2.5634048 5.0060015 7.126855 
512 562.396










    
















    
















    
















    























In [19]:



    


obj = objectives.channel("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=18)



    


















    






0.099485815 2.3148599 4.750007 6.9077325 
512 528.6465










    
















    
















    
















    























In [10]:



    


obj = objectives.neuron("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=18)



    


















    






[0.00690335] [16.094152] [30.655743] [45.56154] 
512 674.91614










    
















    
















    
















    























In [11]:



    


obj = objectives.neuron("mixed4a_pre_relu", 11)
decomposed_vis("conv2d2", obj, W=4*5, W2=18)



    


















    






[-1.7107325] [65.66075] [130.60475] [195.90953] 
512 -107.87578










    
















    
















    
















    























In [14]:



    


obj = objectives.neuron("mixed4a_pre_relu", 476)
decomposed_vis("mixed3b", obj, W=8, W2=32)



    


















    






[0.06190045] [22.612238] [43.084938] [63.457836] 
8 76.5102










    
















    
















    
















    























In [34]:



    


obj = objectives.neuron("mixed4a_pre_relu", 476)
decomposed_vis("mixed3b", obj, W=8, W2=32)



    


















    






[0.04645811] [22.605408] [43.09341] [63.465565] 
128 193.46898










    
















    
















    
















    























In [36]:



    


obj = objectives.neuron("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=32)



    


















    






[0.15653779] [14.630148] [28.307514] [42.193756] 
128 729.6647










    
















    
















    
















    























In [37]:



    


obj = objectives.neuron("mixed4a_pre_relu", 11)
decomposed_vis("conv2d2", obj, W=20, W2=16)



    


















    






[-1.6092606] [91.624405] [181.80292] [271.14542] 
128 15.953862










    
















    
















    
















    























In [11]:



    


obj = objectives.channel("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=32)



    


















    






0.1403646 2.4788575 4.946161 7.074747 
(1, 10, 10, 256) (10, 10, 256) float64
512 597.5731










    
















    
















    
















    























In [13]:



    


obj = objectives.channel("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=18)



    


















    






0.09984322 2.1868072 4.448259 6.4796953 
(1, 10, 10, 256) (10, 10, 256) float64
512 436.8395










    
















    
















    
















    























In [14]:



    


obj = objectives.channel("mixed4a_pre_relu", 11)
decomposed_vis("mixed3a", obj, W=10, W2=16)



    


















    






0.11798098 2.3361197 4.841185 7.0016384 
(1, 10, 10, 256) (10, 10, 256) float64
512 522.15924










    
















    
















    
















    























In [41]:



    


obj = objectives.channel("mixed4a_pre_relu", 1)
decomposed_vis("mixed3b", obj, W=10, W2=32)



    


















    






0.26394105 2.2250974 3.9075282 5.4626045 
128 111.06346










    
















    
















    
















    























In [40]:



    


obj = objectives.channel("mixed4a_pre_relu", 1)
decomposed_vis("mixed3a", obj, W=10, W2=32)



    


















    






0.06347242 2.8805048 5.610231 7.934815 
128 571.7892










    
















    
















    
















    























In [49]:



    


obj = objectives.channel("mixed4a_pre_relu", 1)
decomposed_vis("mixed3a", obj, W=10, W2=32)



    


















    






0.070807226 2.7349432 5.218114 7.41927 
(1, 10, 10, 256) (10, 10, 256) float64










    







InvalidArgumentErrorTraceback (most recent call last)
<ipython-input-49-bea13314e3c5> in <module>()
      1 obj = objectives.channel("mixed4a_pre_relu", 1)
----> 2 decomposed_vis("mixed3a", obj, W=10, W2=32)

<ipython-input-47-1f2a086f5325> in decomposed_vis(layer, obj, W, W2)
     12   ])
     13 
---> 14   imgs = render.render_vis(model, obj, param_f, thresholds=(128,))
     15 
     16   img_grid = np.hstack(np.hstack(imgs[0].reshape([W,W,W2,W2,3])))

/usr/local/lib/python2.7/dist-packages/lucid/optvis/render.pyc in render_vis(model, objective_f, param_f, optimizer, transforms, thresholds, print_objectives, verbose)
     89     try:
     90       for i in range(max(thresholds)+1):
---> 91         loss_, _ = sess.run([loss, vis_op])
     92         if i in thresholds:
     93           vis = t_image.eval()

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict, options, run_metadata)
    898     try:
    899       result = self._run(None, fetches, feed_dict, options_ptr,
--> 900                          run_metadata_ptr)
    901       if run_metadata:
    902         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1133     if final_fetches or final_targets or (handle and feed_dict_tensor):
   1134       results = self._do_run(handle, final_targets, final_fetches,
-> 1135                              feed_dict_tensor, options, run_metadata)
   1136     else:
   1137       results = []

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
   1314     if handle is None:
   1315       return self._do_call(_run_fn, feeds, fetches, targets, options,
-> 1316                            run_metadata)
   1317     else:
   1318       return self._do_call(_prun_fn, handle, feeds, fetches)

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _do_call(self, fn, *args)
   1333         except KeyError:
   1334           pass
-> 1335       raise type(e)(node_def, op, message)
   1336 
   1337   def _extend_graph(self):

InvalidArgumentError: Incompatible shapes: [6,6,256] vs. [1,1,10,10,256]
	 [[Node: mul_103 = Mul[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"](strided_slice_311, mul_103/y)]]
	 [[Node: Adam/update/_46 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_6205_Adam/update", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

Caused by op u'mul_103', defined at:
  File "/usr/lib/python2.7/runpy.py", line 174, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/usr/local/lib/python2.7/dist-packages/ipykernel_launcher.py", line 16, in <module>
    app.launch_new_instance()
  File "/usr/local/lib/python2.7/dist-packages/traitlets/config/application.py", line 658, in launch_instance
    app.start()
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelapp.py", line 477, in start
    ioloop.IOLoop.instance().start()
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/ioloop.py", line 177, in start
    super(ZMQIOLoop, self).start()
  File "/usr/local/lib/python2.7/dist-packages/tornado/ioloop.py", line 888, in start
    handler_func(fd_obj, events)
  File "/usr/local/lib/python2.7/dist-packages/tornado/stack_context.py", line 277, in null_wrapper
    return fn(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/zmqstream.py", line 440, in _handle_events
    self._handle_recv()
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/zmqstream.py", line 472, in _handle_recv
    self._run_callback(callback, msg)
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/zmqstream.py", line 414, in _run_callback
    callback(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/tornado/stack_context.py", line 277, in null_wrapper
    return fn(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelbase.py", line 283, in dispatcher
    return self.dispatch_shell(stream, msg)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelbase.py", line 235, in dispatch_shell
    handler(stream, idents, msg)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelbase.py", line 399, in execute_request
    user_expressions, allow_stdin)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/ipkernel.py", line 196, in do_execute
    res = shell.run_cell(code, store_history=store_history, silent=silent)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/zmqshell.py", line 533, in run_cell
    return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/IPython/core/interactiveshell.py", line 2718, in run_cell
    interactivity=interactivity, compiler=compiler, result=result)
  File "/usr/local/lib/python2.7/dist-packages/IPython/core/interactiveshell.py", line 2828, in run_ast_nodes
    if self.run_code(code, result):
  File "/usr/local/lib/python2.7/dist-packages/IPython/core/interactiveshell.py", line 2882, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-49-bea13314e3c5>", line 2, in <module>
    decomposed_vis("mixed3a", obj, W=10, W2=32)
  File "<ipython-input-47-1f2a086f5325>", line 14, in decomposed_vis
    imgs = render.render_vis(model, obj, param_f, thresholds=(128,))
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/render.py", line 83, in render_vis
    T = make_vis_T(model, objective_f, param_f, optimizer, transforms)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/render.py", line 159, in make_vis_T
    loss = objective_f(T)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 107, in __call__
    return self.objective_func(T)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 86, in <lambda>
    objective_func = lambda T: sum([obj(T) for obj in objs])
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 107, in __call__
    return self.objective_func(T)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 166, in <lambda>
    return lambda T: tf.reduce_mean(T(layer)[batch] * vec)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/math_ops.py", line 859, in binary_op_wrapper
    return func(x, y, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/math_ops.py", line 1091, in _mul_dispatch
    return gen_math_ops.mul(x, y, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gen_math_ops.py", line 4759, in mul
    "Mul", x=x, y=y, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 3414, in create_op
    op_def=op_def)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 1740, in __init__
    self._traceback = self._graph._extract_stack()  # pylint: disable=protected-access

InvalidArgumentError (see above for traceback): Incompatible shapes: [6,6,256] vs. [1,1,10,10,256]
	 [[Node: mul_103 = Mul[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"](strided_slice_311, mul_103/y)]]
	 [[Node: Adam/update/_46 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_6205_Adam/update", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

















In [48]:



    


obj = objectives.channel("mixed3a_pre_relu", 120)
decomposed_vis("conv2d2", obj, W=10, W2=24)



    


















    






0.39785308 8.20059 14.49489 20.791489 
(1, 10, 10, 192) (10, 10, 192) float64










    







InvalidArgumentErrorTraceback (most recent call last)
<ipython-input-48-c5f145992d79> in <module>()
      1 obj = objectives.channel("mixed3a_pre_relu", 120)
----> 2 decomposed_vis("conv2d2", obj, W=10, W2=24)

<ipython-input-47-1f2a086f5325> in decomposed_vis(layer, obj, W, W2)
     12   ])
     13 
---> 14   imgs = render.render_vis(model, obj, param_f, thresholds=(128,))
     15 
     16   img_grid = np.hstack(np.hstack(imgs[0].reshape([W,W,W2,W2,3])))

/usr/local/lib/python2.7/dist-packages/lucid/optvis/render.pyc in render_vis(model, objective_f, param_f, optimizer, transforms, thresholds, print_objectives, verbose)
     89     try:
     90       for i in range(max(thresholds)+1):
---> 91         loss_, _ = sess.run([loss, vis_op])
     92         if i in thresholds:
     93           vis = t_image.eval()

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict, options, run_metadata)
    898     try:
    899       result = self._run(None, fetches, feed_dict, options_ptr,
--> 900                          run_metadata_ptr)
    901       if run_metadata:
    902         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1133     if final_fetches or final_targets or (handle and feed_dict_tensor):
   1134       results = self._do_run(handle, final_targets, final_fetches,
-> 1135                              feed_dict_tensor, options, run_metadata)
   1136     else:
   1137       results = []

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
   1314     if handle is None:
   1315       return self._do_call(_run_fn, feeds, fetches, targets, options,
-> 1316                            run_metadata)
   1317     else:
   1318       return self._do_call(_prun_fn, handle, feeds, fetches)

/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.pyc in _do_call(self, fn, *args)
   1333         except KeyError:
   1334           pass
-> 1335       raise type(e)(node_def, op, message)
   1336 
   1337   def _extend_graph(self):

InvalidArgumentError: Incompatible shapes: [9,9,192] vs. [1,1,10,10,192]
	 [[Node: mul_103 = Mul[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"](strided_slice_311, mul_103/y)]]
	 [[Node: add_3/_49 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_6453_add_3", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

Caused by op u'mul_103', defined at:
  File "/usr/lib/python2.7/runpy.py", line 174, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/usr/local/lib/python2.7/dist-packages/ipykernel_launcher.py", line 16, in <module>
    app.launch_new_instance()
  File "/usr/local/lib/python2.7/dist-packages/traitlets/config/application.py", line 658, in launch_instance
    app.start()
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelapp.py", line 477, in start
    ioloop.IOLoop.instance().start()
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/ioloop.py", line 177, in start
    super(ZMQIOLoop, self).start()
  File "/usr/local/lib/python2.7/dist-packages/tornado/ioloop.py", line 888, in start
    handler_func(fd_obj, events)
  File "/usr/local/lib/python2.7/dist-packages/tornado/stack_context.py", line 277, in null_wrapper
    return fn(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/zmqstream.py", line 440, in _handle_events
    self._handle_recv()
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/zmqstream.py", line 472, in _handle_recv
    self._run_callback(callback, msg)
  File "/usr/local/lib/python2.7/dist-packages/zmq/eventloop/zmqstream.py", line 414, in _run_callback
    callback(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/tornado/stack_context.py", line 277, in null_wrapper
    return fn(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelbase.py", line 283, in dispatcher
    return self.dispatch_shell(stream, msg)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelbase.py", line 235, in dispatch_shell
    handler(stream, idents, msg)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/kernelbase.py", line 399, in execute_request
    user_expressions, allow_stdin)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/ipkernel.py", line 196, in do_execute
    res = shell.run_cell(code, store_history=store_history, silent=silent)
  File "/usr/local/lib/python2.7/dist-packages/ipykernel/zmqshell.py", line 533, in run_cell
    return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/IPython/core/interactiveshell.py", line 2718, in run_cell
    interactivity=interactivity, compiler=compiler, result=result)
  File "/usr/local/lib/python2.7/dist-packages/IPython/core/interactiveshell.py", line 2828, in run_ast_nodes
    if self.run_code(code, result):
  File "/usr/local/lib/python2.7/dist-packages/IPython/core/interactiveshell.py", line 2882, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-48-c5f145992d79>", line 2, in <module>
    decomposed_vis("conv2d2", obj, W=10, W2=24)
  File "<ipython-input-47-1f2a086f5325>", line 14, in decomposed_vis
    imgs = render.render_vis(model, obj, param_f, thresholds=(128,))
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/render.py", line 83, in render_vis
    T = make_vis_T(model, objective_f, param_f, optimizer, transforms)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/render.py", line 159, in make_vis_T
    loss = objective_f(T)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 107, in __call__
    return self.objective_func(T)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 86, in <lambda>
    objective_func = lambda T: sum([obj(T) for obj in objs])
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 107, in __call__
    return self.objective_func(T)
  File "/usr/local/lib/python2.7/dist-packages/lucid/optvis/objectives.py", line 166, in <lambda>
    return lambda T: tf.reduce_mean(T(layer)[batch] * vec)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/math_ops.py", line 859, in binary_op_wrapper
    return func(x, y, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/math_ops.py", line 1091, in _mul_dispatch
    return gen_math_ops.mul(x, y, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gen_math_ops.py", line 4759, in mul
    "Mul", x=x, y=y, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 3414, in create_op
    op_def=op_def)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 1740, in __init__
    self._traceback = self._graph._extract_stack()  # pylint: disable=protected-access

InvalidArgumentError (see above for traceback): Incompatible shapes: [9,9,192] vs. [1,1,10,10,192]
	 [[Node: mul_103 = Mul[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"](strided_slice_311, mul_103/y)]]
	 [[Node: add_3/_49 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_6453_add_3", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

















In [10]:



    


layer = "mixed3b"
obj = objectives.neuron("mixed4a_pre_relu", 476)

vecs = hidden_opt(model, layer, obj, W=8)

param_f = lambda: param.image(32, batch=8*8)
obj = objectives.Objective.sum([
    objectives.direction(layer, vec, batch=n)
    for n, vec in enumerate(vecs.reshape([-1, vecs.shape[-1]]))
])

imgs = render.render_vis(model, obj, param_f)
show(np.hstack(np.hstack(imgs[0].reshape([8,8,32,32,3]))))



    


















    






[0.28909123]
[8.623955]
[17.057825]
[25.491695]
[33.925545]
[42.260395]
[50.6943]
[59.12817]
[67.562]
[75.995865]
512 98.33376










    
















    























In [9]:



    


layer = "mixed3a"
obj = objectives.neuron("mixed4a_pre_relu", 476)

vecs = hidden_opt(model, layer, obj, W=8)

param_f = lambda: param.image(32, batch=8*8)
obj = objectives.Objective.sum([
    objectives.direction(layer, vec, batch=n)
    for n, vec in enumerate(vecs.reshape([-1, vecs.shape[-1]]))
])

imgs = render.render_vis(model, obj, param_f)
show(np.hstack(np.hstack(imgs[0].reshape([8,8,32,32,3]))))



    


















    






512 560.0627










    
















    























In [0]:

Content source: tensorflow/lucid

Similar notebooks:

notebook.community | gallery | about