[R] Facebook AI & University of Notre Dame Propose Multi-Face Pose Estimation Without Face Detection Computer Science |
- [R] Facebook AI & University of Notre Dame Propose Multi-Face Pose Estimation Without Face Detection
- VkResample - real-time Vulkan FFT upscaling
- Most of the courses on Coursera created by the big companies branded like FREE will allow you to access them in the first 7 days and than require payment to continue. We've created list from completely free courses covered by the financial aid, allowing you to learn and get certified for FREE.
- Is there a “right age” to start studying computer science?
- Holiday Entertainment: The Best Books, T.V. Shows, and Movies For Programmers
- Optimal targeting for a sentry gun?
- Why should my son care about English/Social Studies/Literature, they don't advance his coding!
| [R] Facebook AI & University of Notre Dame Propose Multi-Face Pose Estimation Without Face Detection Posted: 16 Dec 2020 01:40 PM PST A new study has proposed a novel real-time six degrees of freedom 3D face pose estimation technique that works without face detection or landmark localization. In the paper Img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation, the University of Notre Dame and Facebook AI research team details their easily trained R-CNN-based model. Here is a quick read: Facebook AI & University of Notre Dame Propose Multi-Face Pose Estimation Without Face Detection The paper img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation is on arXiv, and the team plans to release the implementation on the project GitHub. [link] [comments]  |
| VkResample - real-time Vulkan FFT upscaling Posted: 16 Dec 2020 12:58 PM PST Hello, I am the creator of VkFFT library. In this post I would like to share how it can be used for real-time upscaling of images - VkResample project. Upscaling and supersampling became hot topics in the modern GPU world. There are a lot of options available already, starting from the simplest nearest neighbor algorithm, where each pixel is split in multiple, and coming to Nvidia's DLSS technology, which uses trained neural networks. One of such options is a Fast Fourier Transform (FFT) based upscaling. FFT is an essential algorithm in image processing. It is used for convolution calculations (see: convolution theorem), filtering (high- and low-bandwidth passes), denoising (Wiener filter). FFT can be used as well for Lanczos upsampling, or in other words, for convolutions with sinc window. In the frequency domain, sinc window corresponds to a step-wise function, which is then multiplied with a centered FFT of an image, padded with zeros to the required size and transformed back to the spatial domain to retrieve upscaled image. The main time consuming part, that was previously immensely limiting FFT-based algorithms, was forward and inverse FFTs themselves. The computational cost of them was simply too high to be performed in real-time. However, modern advances in general purpose GPU computing allow for efficient parallelization of FFT, which is done in a form of Vulkan FFT library - VkFFT. It can be used as a part of a rendering process to perform frequency based computations on a frame before showing it to the user. VkResample uses various optimizations available in VkFFT package, such as R2C/C2R mode and native zero padding support, which greatly reduce the amount of memory transfers and computations. With them enabled, it is possible to upscale 2048x1024 image to 4096x2048 in under 2ms on Nvidia GTX 1660Ti GPU. Measured time covers command buffer submission and execution, which include data transfers to the chip, FFT algorithm, modifications in frequency domain and inverse transformation with its own data trasnfers. Support for arbitrary resolutions will be added in one of the next updates of VkFFT. Possible improvements to this algorithm can include: implementing the Discrete Cosine Transform, which is better suited for real-world images; using additional data from previous frames and/or motion vectors; more low-precision tests and optimizations; using deep learning methods in the frequency domain. As of now, VkResample is more of a proof of concept that can be greatly enchanced in the future. Below you can find a collection of screenshots details comparison from Cyberpunk 2077 game upscaled 2x using nearest neighbor method (NN), FFT method (FFT) and rendered in native resolution (Native). Full sized images, source code and executables are available in VkResample GitHub repository. Feel free to ask any questions about the project and VkFFT library (which is now generating better optimized shaders at runtime and has even more supported features than before)! [link] [comments]  |
| Posted: 16 Dec 2020 01:55 PM PST |
| Is there a “right age” to start studying computer science? Posted: 16 Dec 2020 07:25 AM PST I'm 18 and I'm considering start studying code and some languages right now. Data science is something that interests me and I know that it would've been better if I had built basic knowledge about programming earlier in my life, but I only started to admire this area last year. Do you think that "the earlier the better" or that "age is just a number"? Am I at the right age to start coding? [link] [comments]  |
| Holiday Entertainment: The Best Books, T.V. Shows, and Movies For Programmers Posted: 16 Dec 2020 04:37 PM PST Hey guys! It's almost holiday season. So if you're traveling (hopefully safely) and need something on the journey or if you're just staying at home and need something to pass the time, I compiled a list of the best books/movies/TV shows for us computer science enthusiasts. There's some great technical and non-technical picks to get educated, unwind, and learn something new! Feel free to drop your thoughts/opinions and other recommendations! https://blog.thecodex.me/python-culture/ Thanks, Avi [link] [comments]  |
| Optimal targeting for a sentry gun? Posted: 16 Dec 2020 09:08 AM PST I have a sentry gun that needs to shoot anywhere from 2 to 20 randomly places targets on an XY grid. All target coordinates are know ahead of time. The turret starts in a corner and needs to shoot everything as fast as possible. I've been struggling determining the optimal pathing algorithm each time. Is Dijkstra algorithm applicable? Or is there something else that works better for multiple random targets? [link] [comments]  |
| Why should my son care about English/Social Studies/Literature, they don't advance his coding! Posted: 16 Dec 2020 12:31 PM PST My 11yo son is getting extremely into programming. I love that he's interested in it, but the extent he's taking it to is interfering with a lot of other things in life. We're working on that with his therapist. But one thing that's really been suffering has been school. Today he trotted out the "I don't need English, Social Studies, or Literature. It doesn't make me a better coder. It doesn't advance my life goals." I thought I'd ask here for people's thoughts for or against his points. I have my own thoughts on that, of course. My first and most obvious is that: Communication in programming is amazingly important. Writing code so others can understand it is important. Usually those other people are your team-mates. Even if you're going to be a lone-wolf, you need to communicate to future you. Documentation is directly English, but also English lessons translate readable and maintainable code. Why study English, Social Studies, Literature, if you want to get a job coding? Or, why not? [link] [comments]  |
| You are subscribed to email updates from Computer Science: Theory and Application. To stop receiving these emails, you may unsubscribe now. | Email delivery powered by Google |
| Google, 1600 Amphitheatre Parkway, Mountain View, CA 94043, United States | |
No comments:
Post a Comment