Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 3 years ago.
Improve this question
I am building some object detectors with Tensorflow. Really enjoying it.
The most time-intensive part of any project I work on is gathering images and drawing bounding boxes around the relevant classes I want to detect.
In order to get good results, I need to tag at least 100 images for each class - usually more. When the class has a lot of variation - eg a "person" classifier - you need a hell of a lot more.
This brings me to my question, a lot of the famous base networks - MobileNet, VGG-Net, LeNet and the likes come pretrained on a few hundred different classes. So presumably they have pre-labeled/tagged images stored somewhere.
Are these available anywhere for the average developer?
I've found some - e.g from here but they don't seem to contain the labels which would allow me to generate TFRecords to train an object detector.
Tl;dr - Let's say I just want to create a "person" classifier - where can I find labelled training data for this, allowing me to quickly create classes for generating TFRecords/creating an object detector in Tensorflow?
Actually you could make full usage of the pretrained network by transfer learning or fine-tuning on them with your own dataset, it will save a lot of computation time, power.
But if you insist on training from scratch, there are some well-known image datasets like OpenImages, COCO, Kitti, etc.
These datasets contain a lot more classes other than 'person' but you can filter them out when creating tfrecord, so your tfrecord only contains bounding box information about 'person'.
Related
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 1 year ago.
Improve this question
I'm refactoring old code and I want to find all functions that bigger than N lines to refactor it. It would be cool if there was the same tool for finding big classes.
The project has a lot of files so it is not so easy to find large methods manually.
I couldn't find this tool on the Internet, so, maybe you guys know a tool that can help me?
Thanks in advance!
You can find IDE plugins and external tools which will compute cyclomatic complexities of each of your methods to achieve your needs. You can find some tools in this subject
In the PVS-Studio analyzer there is the V553 diagnostic that reports that the size of the function or class exceeds 2000 lines. However, it's not the best idea to rely on the function size. It is because in addition to the length one should also take the function complexity into account. In this case, search for functions with large Cyclomatic complexity may help. For this, there is another V2008 diagnostic in PVS-Studio (note that it's disabled by default).
The tool NDepend can help finding large and complex classes and methods. There are several default rules for that:
Avoid types too big
Avoid methods too big, too complex
Avoid making large methods even larger
Avoid making complex methods even more complex
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 4 years ago.
Improve this question
A human sings a musical tone and wants to convert into a musical instrument playing the same. On processing the .wav file of the recorded human voice using aubio.org, I am able to get the pitch/frequency at different time intervals. But how to get the notes of musical instruments to be played from this data or Is there some tuner that can help me do this ?
For pitch, use the formula that MIDI 69 = 440hz, so (python/pseudo code)
midiNote = round(12 * (log(freq/440.0), 2))) + 69
But you'll find that things can actually be more complex than this for rhythm because. I'd use an already made rhythmic "quantization" package (that's the Google search term you'll want) because determining both the tempo and whether something that is 1.125 beats long should be transcribed as a quarter note or a quarter note tied to a 32nd note will depend a lot on context and it's a big programming job. There are several preexisting toolkits that will do it for you. My own python framework music21 has an audioSearch module that should do the trick for simple pieces -- at least you could take the code there and put it into your own package, since it's pretty standard FFT sample conversion techniques.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 8 years ago.
Improve this question
I should to prepare myself for upcoming task which consist of a lot of graphs.
I need some data (available in free domain) to train myself.
Bigger - is better...
could you suggest some open data resource?
I'll appreciate this.
You can visit http://snap.stanford.edu/data/ . It contains many different kind of network or graph data.
Here is an answer for your could you suggest some open data resource? and not for which consist of a lot of graphs. So, plz, keep it in mind.
Here (data.gov.au) you can find a huge datasets (864!) of a different types in a different formats (txt, csv, xml, ). You will find a Finance, Industry, Geography, etc. datasets.
In other case, if you want some special (and meaningful data, for example, global population density) you can see this (a bit outdated, but usefull) source from readwriteweb.com.
And one more source: "Open Governmental Datasets" - it's worth to see it indeed.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 4 years ago.
Improve this question
Suppose I want to write an application to compose music.
I'd like to feed a set of musical scores of a composer - e.g. Bach's Well tempered clavier' - and the program should prepare new scores in a similar style.
Are algorithms or even libraries known for this task?
WikiPedia provides this page about algorithmic music composition.
You could make something basic using Markov chains. The principle is to first produce some unit of music (a single note, for example) and then, based on the last produced unit, randomly select the next unit.
First, pass through the input music. Each time you see a particular note/other unit of music, simply record in the table what came after it. When you have gone trough the entire input material, you will have a frequency table of which units follow which (After 'A', 'B' appeared 29 times, 'C' appeared 12 times and 'A' appeared twice; after 'B' ... etc).
Now select an initial note. Select the next one randomly according to the frequencies recorded in the table. Repeat until satisfied.
This will probably not yield good results if applied to individual notes, instead try short phrases. Also, the quality will improve if you have access to a large corpus of source music.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 4 years ago.
Improve this question
Can anyone recommend any software/books required to learn and build fractal patterns? I want to also be able to animate the fractal patterns too. Like something off of winamp.
For book on animated fractal Fractals: An Animated Discussion with Edward Lorenz and Benoît Mandelbrot could be what you want.
You might like: The Computational Beauty of Nature
http://mitpress.mit.edu/books/flaoh/cbnhtml/
I am working on a project in java for creating fractals (flame fractals).
The user provides a script file in javascript, that takes care of the interpolation between fractals. Also, as input, one can provide a file that affects an animation in some way,
so this is the way to make an animation change to some music,
see for example
http://www.youtube.com/watch?v=imDPqR9cDL8
I can send you the source code if you provide me with an address.
There's lots of fractal programs out there.
I know that Ultrafractal is pretty popular, but not free.
As far as free ones go, there are programs such a ChaosPro and others. A quick google search will find them for you.