Deep Dream

How to learn own model of DeepDream

https://generateme.wordpress.com/2015/08/12/training-own-dnn-for-deep-dream/

My little tool to help with learning and images preparing and download:
https://github.com/makseq/deepdream_learn_helper

Deep Dream Video

Go here: https://github.com/graphific/DeepDreamVideo

usage example:

-z — zoom of hallucination
-itr — neural network iterations (power of hallucination)
—gpu — gpu number to use
number of guided images from ‘guided’ must be the same as ‘input’!
if ‘caffe’ import error
unpack this (http://m.tka4.org/mpc/caffe.zip) into DeepDreamVideo folder
if ‘libcaffe.so.1.0.0’ not found
download this (http://m.tka4.org/mpc/libcaffe.so.1.0.0) into DeepDreamVideo folder.

GoogleNet layers

 

Layers list

prob
loss3/classifier_zzzz
pool5/drop_7x7_s1
pool5/7x7_s1
inception_5b/output
inception_5b/relu_pool_proj
inception_5b/pool_proj
inception_5b/pool
inception_5b/relu_5x5
inception_5b/5×5
inception_5b/relu_5x5_reduce
inception_5b/5x5_reduce
inception_5b/relu_3x3
inception_5b/3×3
inception_5b/relu_3x3_reduce
inception_5b/3x3_reduce
inception_5b/relu_1x1
inception_5b/1×1
inception_5a/output_inception_5a/output_0_split
inception_5a/relu_pool_proj
inception_5a/pool_proj
inception_5a/pool
inception_5a/relu_5x5
inception_5a/5×5
inception_5a/relu_5x5_reduce
inception_5a/5x5_reduce
inception_5a/relu_3x3
inception_5a/3×3
inception_5a/relu_3x3_reduce
inception_5a/3x3_reduce
inception_5a/relu_1x1
inception_5a/1×1
pool4/3x3_s2_pool4/3x3_s2_0_split
pool4/3x3_s2
inception_4e/output
inception_4e/relu_pool_proj
inception_4e/pool_proj
inception_4e/pool
inception_4e/relu_5x5
inception_4e/5×5
inception_4e/relu_5x5_reduce
inception_4e/5x5_reduce
inception_4e/relu_3x3
inception_4e/3×3
inception_4e/relu_3x3_reduce
inception_4e/3x3_reduce
inception_4e/relu_1x1
inception_4e/1×1
inception_4d/output_inception_4d/output_0_split
inception_4d/output
inception_4d/relu_pool_proj
inception_4d/pool_proj
inception_4d/pool
inception_4d/relu_5x5
inception_4d/5×5
inception_4d/relu_5x5_reduce
inception_4d/5x5_reduce
inception_4d/relu_3x3
inception_4d/3×3
inception_4d/relu_3x3_reduce
inception_4d/3x3_reduce
inception_4d/relu_1x1
inception_4d/1×1
inception_4c/output_inception_4c/output_0_split
inception_4c/output
inception_4c/relu_pool_proj
inception_4c/pool_proj
inception_4c/pool
inception_4c/relu_5x5
inception_4c/5×5
inception_4c/relu_5x5_reduce
inception_4c/5x5_reduce
inception_4c/relu_3x3
inception_4c/3×3
inception_4c/relu_3x3_reduce
inception_4c/3x3_reduce
inception_4c/relu_1x1
inception_4c/1×1
inception_4b/output_inception_4b/output_0_split
inception_4b/output
inception_4b/relu_pool_proj
inception_4b/pool_proj
inception_4b/pool
inception_4b/relu_5x5
inception_4b/5×5
inception_4b/relu_5x5_reduce
inception_4b/5x5_reduce
inception_4b/relu_3x3
inception_4b/3×3
inception_4b/relu_3x3_reduce
inception_4b/3x3_reduce
inception_4b/relu_1x1
inception_4b/1×1
inception_4a/output_inception_4a/output_0_split
inception_4a/output
inception_4a/relu_pool_proj
inception_4a/pool_proj
inception_4a/pool
inception_4a/relu_5x5
inception_4a/5×5
inception_4a/relu_5x5_reduce
inception_4a/5x5_reduce
inception_4a/relu_3x3
inception_4a/3×3
inception_4a/relu_3x3_reduce
inception_4a/3x3_reduce
inception_4a/relu_1x1
inception_4a/1×1
pool3/3x3_s2_pool3/3x3_s2_0_split
pool3/3x3_s2
inception_3b/output
inception_3b/relu_pool_proj
inception_3b/pool_proj
inception_3b/pool
inception_3b/relu_5x5
inception_3b/5×5
inception_3b/relu_5x5_reduce
inception_3b/5x5_reduce
inception_3b/relu_3x3
inception_3b/3×3
inception_3b/relu_3x3_reduce
inception_3b/3x3_reduce
inception_3b/relu_1x1
inception_3b/1×1
inception_3a/output_inception_3a/output_0_split
inception_3a/output
inception_3a/relu_pool_proj
inception_3a/pool_proj
inception_3a/pool
inception_3a/relu_5x5
inception_3a/5×5
inception_3a/relu_5x5_reduce
inception_3a/5x5_reduce
inception_3a/relu_3x3
inception_3a/3×3
inception_3a/relu_3x3_reduce
inception_3a/3x3_reduce
inception_3a/relu_1x1
inception_3a/1×1
pool2/3x3_s2_pool2/3x3_s2_0_split
pool2/3x3_s2
conv2/norm2
conv2/relu_3x3
conv2/3×3
conv2/relu_3x3_reduce
conv2/3x3_reduce
pool1/norm1
pool1/3x3_s2
conv1/relu_7x7
conv1/7x7_s2
input

Examples

Original image

Standard GoogleNet model

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30

Trained on triangles, 165000 iterations

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 9 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/pool

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/pool

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 9 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/pool_proj

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 9 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/5×5

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 20 -itr 20 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/5×5

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 9 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/1×1

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5b/1×1

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_5a/1×1

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l pool4/3x3_s2

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_4b/output

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_4a/output

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_3a/pool

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_3a/1×1

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_3a/5×5

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_3a/5x5_reduce

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l inception_3a/3x3_reduce

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l pool2/3x3_s2

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l conv2/norm2

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l conv2/3x3_reduce

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l pool1/norm1

python 2_dreaming_time.py -i input -o output -it jpg —gpu 0 -z 10 -itr 30 -t /p/deepdream/trainer/data/ -m MYNET_iter_165000.caffemodel -l conv1/7x7_s2

Добавить комментарий

Ваш e-mail не будет опубликован. Обязательные поля помечены *