I'm wondering, how to design a neural network, where the input data can have different shape, as the network has some fixed number of nodes in the input layer.
Typically when I want to train a image classification network for pictures with unknown (various) resolution or when I want to classify a text, with various length.
For example for images I surely can have some preprocessing pipeline which will resize the image, but I can lose some information with it, in the case of text, the "resizing" would be even harder to perform.
Is there any trick, how to design such a network?
Three possibilities come to mind.
The easiest is the zero-padding. Basically, you take a rather big input size and just add zeroes if your concrete input is too small. Of course, this is pretty limited and certainly not useful if your input ranges from a few words to full texts.
Recurrent NNs (RNN) are a very natural NN to choose if you have texts of varying size as input. You input words as word vectors (or embeddings) just one after another and the internal state of the RNN is supposed to encode the meaning of the full string of words. This is one of the earlier papers.
Another possibility is using recursive NNs. This is basically a form of preprocessing in which a text is recursively reduced to a smaller number of word vectors until only one is left - your input, which is supposed to encode the whole text. This makes a lot of sense from a linguistic point of view if your input consists of sentences (which can vary a lot in size), because sentences are structured recursively. For example, the word vector for "the man", should be similar to the word vector for "the man who mistook his wife for a hat", because noun phrases act like nouns, etc. Often, you can use linguistic information to guide your recursion on the sentence. If you want to go way beyond the Wikipedia article, this is probably a good start.
In case of images, instead of asking your NN to recognize what's in your image, you can ask it what's in a particular e.g. 256x256 size part of your picture. You train for that, and use it for certain partly overlapped rolling windows of your whole image. If your to-be-recognized pattern varies in size a lot, you can resize it, and use your NN again.
Related
I'm trying out the Word2Vec tutorial at tensorflow (see here: https://www.tensorflow.org/tutorials/text/word2vec)
While all seems to work fine, the output is somewhat unexpected to me, especially the small cluster in the PCA. The 'closet' words in the embedding dimension also don't make much sense, especially compared to other examples.
Am I doing something (trivially) wrong? Or is this expected?
For completeness, I run this in the nvidia-docker image, but also found similar results running cpu only.
Here is the projected embedding showing the cluster.
There can be various reasons.
One reason is that this is due to the so-called hubness problem of embedding spaces, which is an artifact of the high-dimensional space. Some words end up close to a large part of the space and act as sort of hubs in the nearest neighbor search, so through these words, you can get quickly from everywhere to everywhere.
Another reason might be that the model is just undertrained for this particular word. Word embeddings are typically trained on very large datasets, such that every word appears in sufficiently many contexts. If a word does not appear frequently enough or in too ambiguous contexts, then it also ends up to be similar to basically everything.
I have a bunch of poor quality photos that I extracted from a pdf. Somebody I know has the good quality photo's somewhere on her computer(Mac), but it's my understanding that it will be difficult to find them.
I would like to
loop through each poor quality photo
perform a reverse image search using each poor quality photo as the query image and using this persons computer as the database to search for the higher quality images
and create a copy of each high quality image in one destination folder.
Example pseudocode
for each image in poorQualityImages:
search ./macComputer for a higherQualityImage of image
copy higherQualityImage to ./higherQualityImages
I need to perform this action once.
I am looking for a tool, github repo or library which can perform this functionality more so than a deep understanding of content based image retrieval.
There's a post on reddit where someone was trying to do something similar
imgdupes is a program which seems like it almost achieves this, but I do not want to delete the duplicates, I want to copy the highest quality duplicate to a destination folder
Update
Emailed my previous image processing prof and he sent me this
Off the top of my head, nothing out of the box.
No guaranteed solution here, but you can narrow the search space.
You’d need a little program that outputs the MSE or SSIM similarity
index between two images, and then write another program or shell
script that scans the hard drive and computes the MSE between each
image on the hard drive and each query image, then check the images
with the top X percent similarity score.
Something like that. Still not maybe guaranteed to find everything
you want. And if the low quality images are of different pixel
dimensions than the high quality images, you’d have to do some image
scaling to get the similarity index. If the poor quality images have
different aspect ratios, that’s even worse.
So I think it’s not hard but not trivial either. The degree of
difficulty is partly dependent on the nature of the corruption in the
low quality images.
UPDATE
Github project I wrote which achieves what I want
What you are looking for is called image hashing
. In this answer you will find a basic explanation of the concept, as well as a go-to github repo for plug-and-play application.
Basic concept of Hashing
From the repo page: "We have developed a new image hash based on the Marr wavelet that computes a perceptual hash based on edge information with particular emphasis on corners. It has been shown that the human visual system makes special use of certain retinal cells to distinguish corner-like stimuli. It is the belief that this corner information can be used to distinguish digital images that motivates this approach. Basically, the edge information attained from the wavelet is compressed into a fixed length hash of 72 bytes. Binary quantization allows for relatively fast hamming distance computation between hashes. The following scatter plot shows the results on our standard corpus of images. The first plot shows the distances between each image and its attacked counterpart (e.g. the intra distances). The second plot shows the inter distances between altogether different images. While the hash is not designed to handle rotated images, notice how slight rotations still generally fall within a threshold range and thus can usually be matched as identical. However, the real advantage of this hash is for use with our mvp tree indexing structure. Since it is more descriptive than the dct hash (being 72 bytes in length vs. 8 bytes for the dct hash), there are much fewer false matches retrieved for image queries.
"
Another blogpost for an in-depth read, with an application example.
Available Code and Usage
A github repo can be found here. There are obviously more to be found.
After importing the package you can use it to generate and compare hashes:
>>> from PIL import Image
>>> import imagehash
>>> hash = imagehash.average_hash(Image.open('test.png'))
>>> print(hash)
d879f8f89b1bbf
>>> otherhash = imagehash.average_hash(Image.open('other.bmp'))
>>> print(otherhash)
ffff3720200ffff
>>> print(hash == otherhash)
False
>>> print(hash - otherhash)
36
The demo script find_similar_images also on the mentioned github, illustrates how to find similar images in a directory.
Premise
I'll focus my answer on the image processing part, as I believe implementation details e.g. traversing a file system is not the core of your problem. Also, all that follows is just my humble opinion, I am sure that there are better ways to retrieve your image of which I am not aware. Anyway, I agree with what your prof said and I'll follow the same line of thought, so I'll share some ideas on possible similarity indexes you might use.
Answer
MSE and SSIM - This is a possible solution, as suggested by your prof. As I assume the low quality images also have a different resolution than the good ones, remember to downsample the good ones (and not upsample the bad ones).
Image subtraction (1-norm distance) - Subtract two images -> if they are equal you'll get a black image. If they are slightly different, the non-black pixels (or the sum of the pixel intensity) can be used as a similarity index. This is actually the 1-norm distance.
Histogram distance - You can refer to this paper: https://www.cse.huji.ac.il/~werman/Papers/ECCV2010.pdf. Comparing two images' histograms might be potentially robust for your task. Check out this question too: Comparing two histograms
Embedding learning - As I see you included tensorflow, keras or pytorch as tags, let's consider deep learning. This paper came to my
mind: https://arxiv.org/pdf/1503.03832.pdf The idea is to learn a
mapping from the image space to a Euclidian space - i.e. compute an
embedding of the image. In the embedding hyperspace, images are
points. This paper learns an embedding function by minimizing the
triplet loss. The triplet loss is meant to maximize the distance
between images of different classes and minimize the distance between
images of the same class. You could train the same model on a Dataset
like ImageNet. You could augment the dataset with by lowering the
quality of the images, in order to make the model "invariant" to
difference in image quality (e.g. down-sampling followed by
up-sampling, image compression, adding noise, etc.). Once you can
compute embedding, you could compute the Euclidian distance (as a
substitute of the MSE). This might work better than using MSE/SSIM as a similarity indexes. Repo of FaceNet: https://github.com/timesler/facenet-pytorch. Another general purpose approach (not related to faces) which might help you: https://github.com/zegami/image-similarity-clustering.
Siamese networks for predicting similarity score - I am referring to this paper on face verification: http://bmvc2018.org/contents/papers/0410.pdf. The siamese network takes two images as input and outputs a value in the [0, 1]. We can interpret the output as the probability that the two images belong to the same class. You can train a model of this kind to predict 1 for image pairs of the following kind: (good quality image, artificially degraded image). To degrade the image, again, you can combine e.g. down-sampling followed by
up-sampling, image compression, adding noise, etc. Let the model predict 0 for image pairs of different classes (e.g. different images). The output of the network can e used as a similarity index.
Remark 1
These different approaches can also be combined. They all provide you with similarity indexes, so you can very easily average the outcomes.
Remark 2
If you only need to do it once, the effort you need to put in implementing and training deep models might be not justified. I would not suggest it. Still, you can consider it if you can't find any other solution and that Mac is REALLY FULL of images and a manual search is not possible.
If you look at the documentation of imgdupes you will see there is the following option:
--dry-run
dry run (do not delete any files)
So if you run imgdupes with --dry-run you will get a listing of all the duplicate images but it will not actually delete anything. You should be able to process that output to move the images around as you need.
Try similar image finder I have developed to address this problem.
There is an explanation and the algorithm there, so you can implement your own version if needed.
I have been reading OCR papers such as this one https://arxiv.org/pdf/1704.08628.pdf , and I am have trouble finding out how these datasets are actually generated.
In the linked paper, they use a regressor to predict the start location (a point) and height of a line of text. Then, based on that starting point and height, a second network performs OCR and end of line detection. I realize this is a very simplified explanation, but it follows that their dataset consists (at least in part) of full page text 'images' annotated with where each line begins, and then a transcription of the text on a given line. Alternatively, they could have just used the lower left point of bounding boxes as the start point and the height of the box as the word height (avoiding the need to re-annotate if the data was previously prepared using bounding boxes).
So how is a dataset like this actually created? Looking at other datasets it seems like there is some software that can create XML files containing the ground truths relevant to each image, can someone point me in the right direction? I've been googling around and finding lots of tools for annotating text with sentiment etc and other tools for annotating images for segmentation (for something like a YOLO network), but I'm coming up empty for the creation something like the Maurdoor dataset used in the linked paper.
Thank you
So after submitting this, the related threads window showed me many threads that my googling did not turn up. This http://www.prima.cse.salford.ac.uk/tools software seems to be what I was looking for, but I would still love to hear other ideas.
I need to read data from a very large (~a million entries) text file and am trying to decide which data structure is most appropriate. Each entry in the file contains two integers that represent an edge in a directed graph (the tail and the head vertices), and the vast majority of vertices have at least one outgoing edge. My "naive" solution is to use a vector of vectors, so if the tail vertex was 1 and the head vertex was 2 I'd just do something like graph[1].push_back(2) to read in the entry "1 2". Once the graph is read in I'll be using Kosaraju's algorithm to compute the strongly-connected components, so I figure it will be handy to be able to access each element via the [] operator in constant time.
What are the "typical" choices in terms of data structures in a situation like this? Also, assuming the vector of vectors idea is a bad one, why is it bad? I guess the fact that they vector will need to re-size itself will slow things down, but the number of edges/vertices isn't known until runtime so I'm not sure of a way around that.
Thanks
Do you know number of vertices?
Vector of vectors isn't such bad idea as you think because you can resize the outer vector before reading edges. So copying of the whole graph would be prevented.
As far as I know vector of vectors is good structure for graph. It is often used on olympiads on computer science.
I am working on a project where I take a chess board position (FEN string converted to binary) & it's evaluation score and feed it to a neural network. My aim is to make the neural network differentiate between good and bad positions.
How I encode the position : There are 12 unique pieces in chess i.e pawn, rook, knight, bishop, queen and king for white as well as black. I encode each piece using 4 bits with 0000 denoting an empty square. So the 64 squares are encoded into 256 bits and I use 6 more bits to denote game state like whose turn it is to move, king-castle status, etc.
Problem : Since the input space for chess positions is neither smooth nor uni-modal (one small change in the board position can result in a huge change in the evaluation score), the neural network doesn't learn well. Now, the next logical thing to somehow extract useful features (like material difference, center control, etc) and feed it to the network.
I do not want to hand pick the features as I want the network to learn everything by itself. Therefore I am thinking of extracting features automatically using autoencoders. Is there any better way to accomplish this?
Summary : What is the best way to automatically extract features from a chess board position so that it can be fed into a neural network?
UPDATE : To generate training data, I have modified Stockfish to dump it's evaluation process into a log file. So every new move(position) it considers is written to a file as an FEN string along with it's eval score
Neural networks can give an approximation of any function. The only consideration to do is the dimensionality of the search space, which give constraints to the amount of data you have to get a good approximation.
For a supervised network (you use autoencoders, then I think you use some variant of backpropagation), it's difficult for me to immagine how you think to do the trainig using single positions because you need similar positions in your training set. Maybe your approach is different, but I'm convinced that second strategy (using features) is more promising. I think using positions require a huge amount of data training to get good results.
For features take a look here, and to the classical work of Shannon.
I taked also useful informations from the source code of Crafty.
But you have to extract these informations from the FEN string.
Autoencoders are a way to give a reduction of data (good because increase performances). It seems to be better the use of Pincipal Component Analysys, as reported here.
I hope this can help you.