Deep image prior 'learns' on just one image

Re:Deep image prior 'learns' on just one image by cs702 on Thursday November 30, 2017 @07:28PM (#15820970)

The structure of convolutional neural nets specifies much of the prior knowledge necessary for learning. In other words, the design of these neural nets makes a lot of correct assumptions about the nature of images (stationarity of pixel statistics, locality of pixel dependencies, and so on).

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by dwaltrip on Thursday November 30, 2017 @08:47PM (#15821314)

By structure, we are simply referring to the number of layers, the number of neurons in each layer, and the specific connections between neurons in each pair of neighboring layers, right?

So in this paper, they carefully chose a certain structure, set the weights randomly, and then what happened after that? I understand that they did not then train it with a training data set, but I'm not quite getting what they did with the single distorted image.

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by candiodari on Thursday November 30, 2017 @10:36PM (#15821703)

Well given that it's CNNs, you're leaving out weight sharing.

So by structure you should also include the demand that the prediction of any NxN patch of the image should be roughly equal to the prediction of any other NxN patch of the image.

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by nerfhammer on Thursday November 30, 2017 @07:35PM (#15820997)

is the structure of these CNNs learned or designed?

Do they run some kind of optimizer to learn the optimal CNN structure or does some person sit down and pick structures to include in it?

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by cs702 on Thursday November 30, 2017 @07:37PM (#15821010)

Convolutional layers are designed, by and large, and they're mostly the same everywhere. Yann Le Cun came up with them in the mid-90's, but their academic origins go back to at least the 50's and 60's.

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by nerfhammer on Thursday November 30, 2017 @08:18PM (#15821211)

then the results of this paper are sort of surprising that it's that good, if the structure is just relatively old, hand-crafted and traditional.

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by mvij on Thursday November 30, 2017 @08:08PM (#15821156)

Handcrafted, but auto optimizing is a hot research topic right now with DeepMind et al. Needs to be evolved at deeper level than what they’re doing now so architecture is discovered, not just optimized.

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by fizixer on Thursday November 30, 2017 @09:04PM (#15821376)

By structure, do you mean the architecture of the NN?

Meaning number of hidden layers, nodes per layer, and their connectivity with each other and with the input and output layers?

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by cs702 on Thursday November 30, 2017 @09:08PM (#15821397)

No, the paper means primarily the weight sharing in each kernel filter within each convolutional layer, and the stacking of these layers in deep networks.

[ Parent | on HN ]

Re:Deep image prior 'learns' on just one image by ska on Thursday November 30, 2017 @07:53PM (#15821093)

Which in turn relies on older work.

I wonder if anyone has looked at what Mumford-Shaw implications would look like projected onto a CNN?

[ Parent | on HN ]