52

First up, those images (even the first few) aren't complete trash despite being junk to humans; they're actually finely tuned with various advanced techniques, including another neural network. The deep neural network is the pre-trained network modeled on AlexNet provided by Caffe. To evolve images, both the directly encoded and indirectly encoded images, ...


25

The images that you provided may be unrecognizable for us. They are actually the images that we recognize but evolved using the Sferes evolutionary framework. While these images are almost impossible for humans to label with anything but abstract arts, the Deep Neural Network will label them to be familiar objects with 99.99% confidence. This result ...


13

All answers here are great, but, for some reason, nothing has been said so far on why this effect should not surprise you. I'll fill the blank. Let me start with one requirement that is absolutely essential for this to work: the attacker must know neural network architecture (number of layers, size of each layer, etc). Moreover, in all cases that I examined ...


10

An important question that does not yet have a satisfactory answer in neural network research is how DNNs come up with the predictions they offer. DNNs effectively work (though not exactly) by matching patches in the images to a "dictionary" of patches, one stored in each neuron (see the youtube cat paper). Thus, it may not have a high level view of the ...


8

This problem is called object detection. If you have a trainings set of images with boxed objects you can just train a neural network to directly predict the box. I.e. the output has the same dimension as the input and the NN learns to assign each pixel the probability of belonging to a certain object. If you don't have such a convenient dataset you could ...


7

The similarity of artificial neural networks and the human visual cortex goes very deep, and in many ways the human visual cortex was the inspiration for the techniques we use for the design and implementation of ANNs designed for image recognition. So in that direction, the similarity seems obvious to me. The reverse direction, though, is a question about ...


5

How is it possible that deep neural networks are so easily fooled? Deep neural networks are easily fooled by giving high confidence predictions for unrecognizable images. How is this possible? Can you please explain ideally in plain English? Intuitively, extra hidden layers ought to make the network able to learn more complex classification functions, ...


4

After a bit of research I found something kind of close: Artificially intelligent security cameras are spotting crimes before they happen New surveillance cameras will use computer eyes to find 'pre crimes' by detecting suspicious behaviour and calling for guards CCTV 'fightcams' detect violence 'before it happens' at Dailymail, also check at Telegraph ...


4

One of the Pinterest's white paper about Human Curation and Convnets powering item-to-item recommendationsarxiv describes implementation of convolutional neural network (CNN) based visual features (VGG2014, Faster R-CNN). This demonstrates the effectiveness of it (such image or object representations) which can improve user engagement. The visual features ...


4

You should think of them as different approaches. A deep neural net is a single independent model, whereas ensemble models are ensembles of many independent models. The primary connection between the two is dropout, a particular method of training deep neural nets that's inspired by ensemble methods.


4

Yes! This most certainly can be done. Since you have a labeled dataset, that makes it all the more simple! I would take a look at this project and that should get you where you need to go. The implementation details should be pretty straightforward. Let me know if I can help further.


4

For vision tasks, neural network models almost always include a number of layers that pool and convolute. The convolutions, in particular, are very useful - they can make the model generalize better to inputs and maintain performance when inputs have undergone certain linear transformations (e.g. some scaling or a translation along the x-axis). These ...


3

As a first step, take a look at the introduction chapter from this great review by one of the fathers of deep-learning. For example, from page 9: low-level visual features (like edge detectors) and intermediate-level visual features (like object parts)


3

Actually I'm trying the same thing with the Azure Computer Vision API. Although the API is very good in identifying objects, it has problems identifying specific consumer products (in my experience though). For example it can't really distinguish between two pair of shoes, or two pair of watches. People recommended me using the: Custom Vision Service ...


3

You may play around on an average laptop but training will be very slow and you will be limited on the size of your model. Once you try to build something more serious you will run out of memory very fast. A system with a GPU is recommended if you want to really do things like image recognition. If you buy something I would not go for any GPU with less than ...


3

Can't comment(due to that required 50 rep), but I wanted to make a response to Vishnu JK and the OP. I think you guys are skipping the fact that the neural network only really is saying truly from a programmatic standpoint that "this is most like". For example, while we can list the above image examples as "abstract art", they definitively are most like was ...


3

The neural networks can be easily fooled or hacked by adding certain structured noise in image space (Szegedy 2013, Nguyen 2014) due to ignoring non-discriminative information in their input. For example: Learning to detect jaguars by matching the unique spots on their fur while ignoring the fact that they have four legs.2015 So basically the high ...


3

The approach you listed here is not really an approach, this is very very vague idea of how someone can achieve some task. You basically told we have an algorithm f(image) = result and there can be infinite amount of real approaches to solve this. In majority of CNN approaches the image travels through a convolution/pooling layers which reduces the ...


3

Your first approach doesn't make any sense to me. After all, you are not just interested in the centre pixel are you? And if you have two nodes for every pixel, what are those two nodes encoding? For the probability of water you just need one. So clearly approach two. I would just use 0.5 as cutoff. Using a higher or lower cutoff only makes sense, if ...


3

I am not into the field of super resolution but I think this question applies to general neural network construction. Usually you try to solve a classification problem or a regression problem with your neural network. For classification you try to predict probabilities that a specific output corrensponds to a specific class. Therefore every output value ...


3

StarCraft II is a real time strategy game that combines fast paced micro actions with the need for high level planning and execution. StarCraft II being a popular game with millions of users it proceeds that defeating top players becomes a meaningful and measurable long term objective in AI research. Computer games provide a compelling solution to the ...


3

While training a GAN, 2 models are used. A generator and a discriminator. This training process usually takes hours (or days) to complete. This is an offline process and is not happening in the browser. The pict file is the pre-trained model that has been imported to deeplearn.js for inference. The example you have linked to above accepts a sketch drawn in ...


3

If deep learning is what you are trying to use here, you should keep in mind that the real intent behind deep learning is to learn a probability distribution, which means that if you were to use a deep learning model to "rotate" images, you can only do it on a specific class of images (e.g. faces, cats, etc...). If that's your goal, generative models are ...


3

— Stereoscopic Synthesis — The generation of an image that would likely appear in the right eye of a head from which you already have an image from the left eye (or vice versa) is too complex to expect simple convolution (linear matrix transformation) to achieve a reasonable result. You are correct that rotation is not the correct description, simply ...


3

There are a few. Sloth LabelBox RectLabel LabelMe OpenCV also has some facilitation for annotating. Annotation is the more common name used in software suites for tools that facilitate adding labels to images and frames of movies because. Simple categorization is of limited use in the real world, especially with moving pictures where action may be ...


3

Great question, and one that I think we could have done a better job of answering in the paper. Essentially, the pose matrix of each capsule is set up so that it could learn to represent the affine transformation between the object and the viewer, but we are not restricting it to necessarily do that. So we talk about the output of a capsule as though it is ...


3

Depending on kind and amount of data you posess, there are few approaches that you might consider. Marking target objects on dataset and training CNN that returns coordinates of target object. In this case, remember that it is usually faster when training data ROIs have their coordinates relative to image size. Use some kind of focus mechanism, like spatial ...


3

I think you are slightly confusing 2 problems. 1 being classification of meta visual elements and the other being the visual system itself. Our visual system, when it comes to processing information, has had billions of years of iteration(training), so that at birth(and before), we are already tuned for the processing of visual stimuli, as well as have the ...


3

I've seen this too many times - it's not a problem with your network, it's a problem with matplotlib and how it displays the image. You are probably trying to display a float with range $<0, 255>$. When matplotlib sees float type as input, it assumes a range of $<0, 1>$, and thresholds everything outside of that range, and the results you can see....


3

Facial classification and tracking is easier compared to tracking any other motion. This is due to the fact that the face have a large number of easily identifiable features. Facial tracking is an additional layer on top of facial detection. Facial detection works by finding characteristics such as the cheekbones, chin, nose, eyes etc. These features are ...


Only top voted, non community-wiki answers of a minimum length are eligible