I am trying to train a CNN network on optical flow. And I need to generate my one frames with the .flo ground truth. However, I don't know how to generate the flo files. So, my question is: How can I generate a ground true optical flow file .flo?
Generating ground-thruth by your own is almoust imposibble. But you can used the ground-thruth provided by the Middlebury, SINTEL or KITTI. In order to estimate your own optical flow fields i would suggest to used OpenCV 3.0 or higher and download the contribution git repo. The repo includes a function to save optical flow fields in the .flo format.
Related
Even though I am quite familiar with the concepts of Machine Learning & Deep Learning, I never needed to create my own dataset before.
Now, for my thesis, I have to create my own dataset with images of an object that there are no datasets available on the internet(just assume that this is ground-truth).
I have limited computational power so I want to use YOLO, SSD or efficientdet.
Do I need to go over every single image I have in my dataset by my human eyes and create bounding box center coordinates and dimensions to log them with their labels?
Thanks
Yes, you will need to do that.
At the same time, though the task is niche, you could benefit from the concept of transfer learning. That is, you can use a pre-trained backbone in order to help your model to learn faster/achieve better results/need fewer annotations example, but you will still need to annotate the new dataset on your own.
You can use software such as LabelBox, as a starting point, it is very good since it allows you to output the format in Pascal(VOC) format, YOLO and COCO format, so it is a matter of choice/what is more suitable for you.
I'm looking for best practices and performance-guided recommendations for recomputing a model's volume when it's missing from the source file. This is in the context of a web application I am working to build that enables:
Uploading 3D models in a variety of file formats
Interacting with these models using the AutoDesk Viewer
Displaying mass properties, eg volume and surface area, alongside the viewer (subject of this post)
Background
Some file formats have very reliable volume information that is computed and written to the file by the authoring application. For these files, we can access volume as a property via AutoDesk Viewer.
Other formats, however, do not carry volume information - at least not in a manner that is openly accessible using tools other than the authoring application (prime example here is SolidWorks). This leaves us with a giant gap to fill - we need to recompute the model's volume using what's in the file.
Known Workarounds and Options
AutoDesk published a blog post detailing an approach for approximating model volume using triangles of the model inside the viewer. I think it's an ideal solution for use cases that can afford to trade accuracy for a bump in performance - and it centers everything in the viewer making development and subsequent maintenance simpler. This application, however, cannot rely on such approximations. I'm left reviewing options for leveraging the AutoDesk Design Automation API to:
Spin up an instance of Inventor
Load the model file
Rely on iLogic to trigger a re-computation of the model's part properties (perhaps like this?)
Push that data back to my web application
Where I Need Help
My understanding is that an AppBundle and Activity are defined ahead of time and then every uploaded model would be submitted as a work item.
I am hoping for guidance in:
whether this is the only approach or whether there are other options worth considering
how best to orchestrate the end-to-end process from an order of operations/workflow standpoint to maximize performance
Current Thinking
For example, I'm thinking that my first step after the source file is uploaded is to immediately initialize two parallel processes: the first to translate the source file for the viewer, the second to spin up Inventor and trigger the related downstream process to get volume.
The other option I'm considering is handling all of the work in Inventor - and pushing out an SVF file to the viewer that's enriched with volume data. The advantage of this approach is that my frontend will have only one source for volume data, (it will be in the enriched SVF no matter whether it was supplied in the original file or not).
In an ideal world I'd be able to only invoke the Design Automation API when volume data is missing from the source file - but I'd only know that after translating the file and bringing it back to the viewer. Given that many of our files are created in SolidWorks and other high-end proprietary CAD platforms, my working hypothesis is that we'll be needing to fill in volume gaps more often than not.
Your understanding is correct:
appbundle is simply a collection of files (binaries, data) encapsulating a specific Inventor/Revit/3dsMax/AutoCad plugin
activity is a kind of a job template specifying which application should be invoked, which appbundle should be loaded into the application, what inputs will be provided to the job, and what outputs will be generated
work item is then a specific instance of a job, binding the activity inputs and outputs to specific URLs
There is currently no other way to access the Design Automation functionality other then using these 3 types of entities.
I would suggest the following:
wherever possible, use the Design Automation for Inventor to compute the precise areas/volumes
for file formats that cannot be imported into Inventor or any other Design Automation engine, you could use tools like https://github.com/petrbroz/forge-convert-utils to parse the SVF and compute (a very rough estimate of) the area/surface from the triangular meshes; however, this will be quite computationally expensive, and imprecise
Can machine learning be used to transform/modifiy a list of numbers.
I have many pairs of binary files read from vehicle ECUs, an original or stock file before the vehicle was tuned, and a modified file which has the engine parameters altered. The files are basically lists of little or big endian 16 bit numbers.
I was wondering if it is at all possible to feed these pairs of files into machine learning, and for it to take a new stock file and attempt to transform or tune that stock file.
I would appreciate it if somebody could tell me if this is something which is at all possible. All of the examples I've found appear to make decisions on data rather than do any sort of a transformation.
Also I'm hoping to use azure for this.
We would need more information about your specific problem to answer. But, supervised machine learning can take data with a lot of inputs (like your stock file, perhaps) and an output (say a tuned value), and learn the correlations between those inputs and output, and then be able to predict the output for new inputs. (In machine learning terminology, these inputs are called "features" and the output is called a "label".)
Now, within supervised machine learning, there is a category of algorithms called regression algorithms. Regression algorithms allow you to predict a number (sounds like what you want).
Now, the issue that I see, if I'm understanding your problem correctly, is that you have a whole list of values to tune. Two things:
Do those values depend on each other and influence each other? Do any other factors not included in your stock file affect how the numbers should be tuned? Those will need to be included as features in your model.
Regression algorithms predict a single value, so you would need to build a model for each of the values in your stock file that you want to tune.
For more information, you might want to check out Choosing an Azure Machine Learning Algorithm and How to choose algorithms for Microsoft Azure Machine Learning.
Again, I would need to know more about your data to make better suggestions, but I hope that helps.
Is there a simple way (e.g. without modifying caffe code) to load wights from multiple pretrained networks into one network? The network contains some layers with same dimensions and names as both pretrained networks.
I am trying to achieve this using NVidia DIGITS and Caffe.
EDIT: I thought it wouldn't be possible to do it directly from DIGITS, as confirmed by answers. Can anyone suggest a simple way to modify the DIGITS code to be able to select multiple pretrained networks? I checked the code a bit, and thought the training script would be a good place to start, but I don't have in-depth knowledge of Caffe, so I'm not sure what the best/quickest way to achieve this would be.
As Shai suggested, there was no way of doing this, so I decided to clone the official repository and make the appropriate changes. I changed the code so that multiple pretrained networks can be loaded by using a colon as separator.
I created a pull request on the official repository and my changes were then merged with the main branch of DIGITS, meaning it is now possible to use this functionality in DIGITS.
AFAIK there is no straight forward way of doing so.
However, you can use net surgery to load the pretrained models and manually assign their weights to the target net. Once you have a single net with all the weights initialized according to the various pretrained models, you can save it and use it as a single pretrained model for the rest of your work.
I have three different Stata files (each for three different years) and I want to estimate a fixed effects regression. My guess is that I need to merge those files in order to test my regression, but how do I do it? How do I give the time identification for the same variable in each of these files?
Typically, you don't merge (put the files side by side) such files, but append (put them on top of one another) them. Typically, the year or wave variable is already included, but when that is not the case you need to generate them before you merge the files. For more, just type in Stata help merge, help append, and help generate.
Preparing datasets should be exactly documented, so using the GUI is not the way to do this. Instead, you should do this using a .do file. For a good introduction on how to do good and reproducible research with Stata, see:
Long, J. S. (2009). The workflow of data analysis using Stata. College Station, TX: Stata Press.