Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 3 years ago.
Improve this question
I am trying to implement a Convolutional Neural Network (CNN) model to classify hand gestures. Dataset is not readily available and hence I need to prepare it.
How should i prepare the dataset? Should the images I capture contain objects other than the hand or only the hand? Which will give me an accurate model that will work accurately despite of background and other objects in the frame?
Good Dataset for your problem:
You should consider involving different backgrounds and objects in background.
Following links might help you:
https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html
https://www.quora.com/Computer-Vision-What-is-the-best-way-to-collect-Train-and-Test-data-images-for-object-recognition-job
here is an example:
http://cims.nyu.edu/~tompson/NYU_Hand_Pose_Dataset.htm
it containing other images would just mean you have to implement something in your pipeline to isolate the hand. i would recommend having only the hand in the images so you can just start modelling on the images right away.
a lot of cnn architectures in this area using multi-resolution CNNs. so in your data preparation just make multiple resolutions and feed to a multi input CNN. you can make this using Keras functional API. low res images are fine for differentiating between certain very different poses and the higher res can focus on small differences.
obviously, standard data augmentation is not that suitable for hand pose. stuff like mirroring or changing the angle could make your data unsuitable for the given label. so be a bit more conservative with your data augmentation if you don't have that much.
Related
Closed. This question needs details or clarity. It is not currently accepting answers.
Want to improve this question? Add details and clarify the problem by editing this post.
Closed 2 years ago.
Improve this question
I'm learning image classification with Pytorch. I found some papers code use 'CenterCrop' to both train set and test set，e.g. Resize to larger size，then apply CenterCrop to obtain a smaller size. The smaller size is a general size in this research direction.
In my experience, I found apply CenterCrop can get a significant improvement(e.g. 1% or 2%) on test, compare to without CenterCrop on test set.
Because it is used in the top conference papers, confused me. So, Should CenterCrop be used to test set this count as cheating？ In addition, should I use any data augmentation to test set except 'Resize' and 'Normalization'?
Thank you for your answer.
That is not cheating. You can apply any augmentation as long as the label is not used.
In image classification, sometimes people use a FiveCrop+Reflection technique, which is to take five crops (Center, TopLeft, TopRight, BottomLeft, BottomRight) and their reflections as augmentations. They would then predict class probabilities for each crop and average the results, typically giving some performance boost with 10X running time.
In segmentation, people also use similar test-time augmentation "multi-scale testing" which is to resize the input image to different scales before feeding it to the network. The predictions are also averaged.
If you do use such kind of augmentation, do report them when you compare with other methods for a fair comparison.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 3 years ago.
Improve this question
I am working on image segmentation and object detection and I think they do the same thing (they both localize and recognize objects in the raw picture). Is there any benefit in using object detection at all cause deeplab_V3+ do the job with better performance than any other object detection algorithms?
you can look at deeplab_V3+ demo in here
In object detection, the method localizes and classifies the object in the image based on bounding box coordinates. However, in image segmentation, the model also detects the exact boundaries of the object, which usually makes it a bit slower. They both have their own uses. In many applications (e.g. face detection), you only want to detect some specific objects in images and don't necessarily care about the exact boundaries of them. But in some applications (e.g. medical images), you want the exact boundaries of a tumor for example. Also we can consider the process of preparing the data for these tasks:
classification: we only provide a label for each image
localization: we provide a bounding box (4 elements) for each image
detection: we should provide a bounding box and a label for each object
segmentation: we need to define the exact boundaries of each object (semantic segmentation)
So for segmentation, more work is required both in providing the data and in training a (encoder-decoder) model, and it depends on your purpose of the task.
Closed. This question needs details or clarity. It is not currently accepting answers.
Want to improve this question? Add details and clarify the problem by editing this post.
Closed 2 years ago.
Improve this question
I am interested in knowing the importance of data augmentation(rotation at various angles, flipping the images) while providing a dataset to a Machine Learning problem.
Whether it is really needed? Or the CNN networks using will handle that as well no matter how different the data are transformed?
So I took a classification task with 2 classes to conclude some results
Arrow shapes
Circle shapes
The idea is to train the shapes with only one orientation(I have taken arrows pointing right) and check the model with a different orientation(I have taken arrows pointing downwards) which is not at all given during the training stage.
Some of the samples used in Training
Some of the samples used in Testing
This is the entire dataset I am using in for creating a tensorflow model.
https://bitbucket.org/akhileshmalviya/samples/src/bab50b85d826?at=master
I am wondering with the results I got,
(i) Except a few downward arrows all others are getting predicted correctly as arrow. Does it mean data augmentation is not at all needed?
(ii) Or is this the right use case I have taken to understand the importance of data augmentation?
Kindly share your thoughts, Any help could be really appreciated!
Data augmentation is a data-depended process.
In general, you need it when your training data is complex and you have a few samples.
A neural network can easily learn to extract simple patterns like arcs or straight lines and these patterns are enough to classify your data.
In your case data augmentation can barely help, the features the network will learn to extract are easy and highly different from each other.
When you, instead, have to deal with complex structures (cats, dogs, airplanes, ...) you can't rely on simple features like edges, arcs, etc..
Instead, you have to show to your network that the instances you're trying to classify got an high variance and that the features extracted can be combined in a lot of different ways for the same subject.
Think about a cat: it can be of any color, the picture can be taken in different light conditions, its whole body can be in any position, the picture could be taken with a certain orientation...
To correctly classify instances so different, the network must learn to extract robust features that could be learned only after seeing a lot of different inputs.
In your case, instead, simple features can completely discriminate your input, thus any sort of data augmentation could help by just a little bit.
The task you are solving can be easily solved without any NN and even without machine learning.
Just because the problem is so simple it does not really matter whether you do a data augmentation or not. The need for data augmentation is task specific and depends on many things:
how easy is to augment the data with preserving the ability to correctly mark the class. For image, sounds which we used to see/hear it is not a problem (we know that adding small noise to the sound does not change the meaning, rotating the lizard is still a lizard). For other things augmenting without preserving the class/value is hard (for example in Go, randomly adding a stone can change the value of the position dramatically)
does the augmented data is drawn from the same distribution you care about. Adding random stones to Go does not work, but rotating flipping the board works and preserves distribution. But for example in a racing king game (variant of chess) it will not help. You can't flip the position (left <-> right), the evaluation stays the same, but it will never happen in real game and therefore drawn from different distribution and useless
how much data do you have and how expressive is your model. The more parameters you model have, the bigger the chance of overfitting and the more is your need for data. If you train a linear regression in n dims, you will have n + 1 params. You do not really need to augment this. Also if you already have 10bln data points, the augmentation is probably will not be helpful.
how expensive the augmentation procedure. For rotating/scaling the image it is very cheap, but for other augmentation it can be computationally expensive
something else that I forgot.
Closed. This question needs details or clarity. It is not currently accepting answers.
Want to improve this question? Add details and clarify the problem by editing this post.
Closed 7 years ago.
Improve this question
HOG is popular in human detection. Can it be used for detecting objects like cup in the image for example.
I am sorry for not asking programming question, but I mean to get the idea if i can use hog to extract object features.
According to my research I have dont for few days I feel yes but I am not sure.
Yes, HOG (Histogram of Oriented Gradients) can be used to detect any kind of objects, as to a computer, an image is a bunch of pixels and you may extract features regardless of their contents. Another question, though, is its effectiveness in doing so.
HOG, SIFT, and other such feature extractors are methods used to extract relevant information from an image to describe it in a more meaningful way. When you want to detect an object or person in an image with thousands (and maybe millions) of pixels, it is inefficient to simply feed a vector with millions of numbers to a machine learning algorithm as
It will take a large amount of time to complete
There will be a lot of noisy information (background, blur, lightning and rotation changes) which we do not wish to regard as important
The HOG algorithm, specifically, creates histograms of edge orientations from certain patches in images. A patch may come from an object, a person, meaningless background, or anything else, and is merely a way to describe an area using edge information. As mentioned previously, this information can then be used to feed a machine learning algorithm such as the classical support vector machines to train a classifier able to distinguish one type of object from another.
The reason HOG has had so much success with pedestrian detection is because a person can greatly vary in color, clothing, and other factors, but the general edges of a pedestrian remain relatively constant, especially around the leg area. This does not mean that it cannot be used to detect other types of objects, but its success can vary depending on your particular application. The HOG paper shows in detail how these descriptors can be used for classification.
It is worthwhile to note that for several applications, the results obtained by HOG can be greatly improved using a pyramidal scheme. This works as follows: Instead of extracting a single HOG vector from an image, you can successively divide the image (or patch) into several sub-images, extracting from each of these smaller divisions an individual HOG vector. The process can then be repeated. In the end, you can obtain a final descriptor by concatenating all of the HOG vectors into a single vector, as shown in the following image.
This has the advantage that in larger scales the HOG features provide more global information, while in smaller scales (that is, in smaller subdivisions) they provide more fine-grained detail. The disadvantage is that the final descriptor vector grows larger, thus taking more time to extract and to train using a given classifier.
In short: Yes, you can use them.
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 3 years ago.
Improve this question
I'm trying to develop a system, which recognizes various objects present in an image based on their primitive features like texture, shape & color.
The first stage of this process is to extract out individual objects from an image and later on doing image processing on each one by one.
However, segmentation algorithm I've studied so far are not even near perfect or so called Ideal Image segmentation algorithm.
Segmentation accuracy will decide how much better the system responds to given query.
Segmentation should be fast as well as accurate.
Can any one suggest me any segmentation algorithm developed or implemented so far, which won't be too complicated to implement but will be fair enough to complete my project..
Any Help is appreicated..
A very late answer, but might help someone searching for this in google, since this question popped up as the first result for "best segmentation algorithm".
Fully convolutional networks seem to do exactly the task you're asking for. Check the paper in arXiv, and an implementation in MatConvNet.
The following image illustrates a segmentation example from these CNNs (the paper I linked actually proposes 3 different architectures, FCN-8s being the best).
Unfortunately, the best algorithm type for facial recognition uses wavelet reconstruction. This is not easy, and almost all current algorithms in use are proprietary.
This is a late response, so maybe it's not useful to you but one suggestion would be to use the watershed algorithm.
beforehand, you can use a generic drawing(black and white) of a face, generate a FFT of the drawing---call it *FFT_Face*.
Now segment your image of a persons face using the watershed algorithm. Call the segmented image *Water_face*.
now find the center of mass for each contour/segment.
generate an FFT of *Water_Face*, and correlate it with the *FFT_Face image*. The brightest pixel in resulting image should be the center of the face. Now you can compute the distances between this point and the centers of segments generated earlier. The first few distances should be enough to distinguish one person from another.
I'm sure there are several improvements to the process, but the general idea should get you there.
Doing a Google search turned up this paper: http://www.cse.iitb.ac.in/~sharat/papers/prim.pdf
It seems that getting it any better is a hard problem, so I think you might have to settle for what's there.
you can try the watershed segmentation algorithm
also you can calculate the accuracy of the segmentation algorithm by the qualitative measures