layer = ‘mixed4d_3x3_bottleneck_pre_relu’
T(layer)[:,:,:,142] + T(layer)[:,:,:,8]
The code we wrote was only for easily running a photo on a specific layer and every channel in that layer. We modified the render_deapdream function so that it returned the image to be saved into the correct directory.
def render_deepdream(...)
...
return PIL.Image.fromarray(np.uint8(np.clip(img/255.0, 0, 1)*255))
image_name = 'insert_image_path'
layer = 'mixed4d_3x3_bottleneck_pre_relu'
new_file_path = './'+layer+'/'
for i in range(1, 84):
img0 = PIL.Image.open(image_name)
img0 = np.float32(img0)
deep_dream_image = render_deepdream(T(layer)[:,:,:,i], img0)
deep_dream_image.save(new_file_path+str(i)+'.jpeg')
We included two of those tests in this repo. One was for mized4d_3x3_bottleneck_pre_relu and another was for mixed4b_3x3_bottleneck_pre_relu.
We used the following image and you will see the images laid out in the readmes of the respective layers.
We also had some intermediate results before we settled on eyes for our final project. Some of those results can be found in the intermediate results folder.