Showing posts with label AI. Show all posts
Showing posts with label AI. Show all posts

Tuesday, February 07, 2023

Synthesia is an AI video creation platform.

 https://www.synthesia.io/


Is Synthesia.io free? You can make a free demo video to test the AI generator (limited to 200 words). Simply upload or type in your script, and get your video.If you like the results, you can upgrade to the Personal pricing plan ($30/month) or a custom Corporate plan.

Tuesday, September 13, 2022

iniVation - DAVIS 346 Neuromorphic Event camera.

A high-precision camera sensor manufactured by Swiss company iniVation, which develops neuromorphic vision systems.

Employs a patented technology that works like the human retina, significantly reducing power consumption, data rate and computational requirements compared to conventional camera sensors, while providing a dynamic range of 120dB and latency of 1ms. Has less and higher performance.

It is expected to be used in the following fields:

surveillance and ambient sensing / high-speed robots / FA applications (factory automation) / microscopy / motion analysis (humans and animals) / fluid dynamics / sleep research, chronobiology / Fluorescence Imaging / Particle Tracking



Product Specifications:

  • 346 x 260 resolution

  • Simultaneous frame and event output

  • H 40 x W 60 x D 25 mm

  • 100 g without lens

  • CE certified

  • Color or mono

Device comes with:

  • Micro-USB 3.0 cable

  • CS-mount lens

  • 1/4" mount





The camera does not output frames of intensity images. However, similar intensity images can be reconstructed from the event output by their DV software. 


There is support for multi-camera time synchronization via daisy chain connection and external event injection.









iniVation website: https://inivation.com/




CED: Color Event Camera Dataset
CED is the first Color Event Camera Dataset, featuring 50 minutes of footage with both color frames and color events from the Color-DAVIS346. Event cameras have several advantages over conventional cameras: high dynamic range, low latency and immunity to motion blur. We include sequences that showcase these attractive properties, such as high-speed, low-light etc. All the data are released as binary (rosbag) files.



Thursday, May 13, 2021

Intel is using machine learning to make GTA V look incredibly realistic



Enhancing Photorealism Enhancement


Intel ISL

Enhancing Photorealism Enhancement
Stephan R. Richter, Hassan Abu AlHaija, and Vladlen Koltun


We present an approach to enhancing the realism of synthetic images. The images are enhanced by a convolutional network that leverages intermediate representations produced by conventional rendering pipelines. The network is trained via a novel adversarial objective, which provides strong supervision at multiple perceptual levels. We analyze scene layout distributions in commonly used datasets and find that they differ in important ways. We hypothesize that this is one of the causes of strong artifacts that can be observed in the results of many prior methods. To address this we propose a new strategy for sampling image patches during training. We also introduce multiple architectural improvements in the deep network modules used for photorealism enhancement. We confirm the benefits of our contributions in controlled experiments and report substantial gains in stability and realism in comparison to recent image-to-image translation methods and a variety of other baselines.





Friday, January 01, 2021

Synthesia - AI generated Actors, Text to video

Synthesia is powerful and easy way to create AI videos from text The future of synthetic media is here,   http://www.synthesia.io


Feed it text and it will generate a photorealistic human avatar actor that will recite the script. 



Wednesday, August 05, 2020

Netflix Research


Just stumbled on this cool website where Netflix openly posts about some of the research they are doing. 


Video Encoding and Quality
https://research.netflix.com/research-area/video-encoding-and-quality

AI and Machine Learning

Streaming

Tuesday, July 14, 2020

High-Fidelity Generative Image Compression



https://hific.github.io/

We combine (GAN) Generative Adversarial Networks with learned compression to obtain a state-of-the-art generative lossy compression system. In the paper, we investigate normalization layers, generator and discriminator architectures, training strategies, as well as perceptual losses. In a user study, we show that our method is preferred to previous state-of-the-art approaches even if they use more than 2× the bitrate.






Sunday, May 12, 2019

new A.I. camera that can spot you from 28 miles away



A new camera can photograph you from 45 kilometers away
Developed in China, the lidar-based system can cut through city smog to resolve human-sized features at vast distances.
by Emerging Technology from the arXiv
May 3, 2019

Long-distance photography on Earth is a tricky challenge. Capturing enough light from a subject at great distances is not easy. And even then, the atmosphere introduces distortions that can ruin the image; so does pollution, which is a particular problem in cities. That makes it hard to get any kind of image beyond a distance of a few kilometers or so (assuming the camera is mounted high enough off the ground to cope with Earth’s curvature).

But in recent years, researchers have begun to exploit sensitive photodetectors to do much better. These detectors are so sensitive they can pick up single photons and use them to piece together images of subjects up to 10 kilometers (six miles) away.

Nevertheless, physicists would love to improve even more. And today, Zheng-Ping Li and colleagues from the University of Science and Technology of China in Shanghai show how to photograph subjects up to 45 km (28 miles) away in a smog-plagued urban environment. Their technique uses single-photon detectors combined with a unique computational imaging algorithm that achieves super-high-resolution images by knitting together the sparsest of data points.

The new technique is relatively straightforward in principle. It is based on laser ranging and detection, or lidar—illuminating the subject with laser light and then creating an image from reflected light.

The big advantage of this kind of active imaging is that the photons reflected from the subject return to the detector within a specific time window that depends on the distance. So any photons that arrive outside this window can be ignored.

This “gating” dramatically reduces the noise created by unwanted photons from elsewhere in the environment. And it allows lidar systems to be highly sensitive and distance specific.

To make the new system even better in urban environments, Zheng-Ping and co use an infrared laser with a wavelength of 1550 nanometers, a repetition rate of 100 kilohertz,  and a modest power of 120 milliwatts. This wavelength makes the system eye-safe and allows the team to filter out solar photons that would otherwise overwhelm the detector.

The researchers send and receive these photons through the same optical apparatus—an ordinary astronomical telescope with an aperture of 280 mm. The reflected photons are then detected by a commercial single-photon detector. To create an image, the researchers scan the field of view using a piezo-controlled mirror that can tilt up, down, and side to side.

In this way, they can create two-dimensional images. But by changing the gating timings, they can pick up photons reflected from different distances to build a 3D image.

The final advance the team has made is to develop an algorithm that knits an image together using the single-photon data. This kind of computational imaging has advanced in leaps and bounds in recent years, allowing researchers to create images from relatively small sets of data.

The results speak for themselves. The team set up the new camera on the 20th floor of a building on Chongming Island in Shanghai and pointed it at the Pudong Civil Aviation Building across the river, some 45 km away.

single pixel resolution imaging

Conventional images taken through the telescope show nothing other than noise. But the new technique produces images with a spatial resolution of about 60 cm, which resolves building windows. “This result demonstrates the superior capability of the near-infrared single-photon LiDAR system to resolve targets through smog,” say the team.

That’s also significantly better than the conventional diffraction limit of 1 meter at 45 km, and certainly better than other recently developed algorithms. The image here shows the potential of the technique in images taken in daylight from a distance of 21 km. ”Our results open a new venue for high-resolution, fast, low-power 3D optical imaging over ultralong ranges,” say Zheng-Ping and co.

That’s interesting work that has a wide range of applications. The team mention remote sensing, airborne surveillance, and target recognition and identification. Indeed, the entire device is about the size of a large shoebox and so is relatively portable.

And Zheng-Ping and co say it can be significantly improved. “Our system is feasible for imaging at a few hundreds of kilometers by refining the setup, and thus represents a significant milestone towards rapid, low-power, and high-resolution LiDAR over extra-long ranges,” they say.

So keep smiling—they may be watching.

Ref: arxiv.org/abs/1904.10341 : Single-Photon Computational 3D Imaging at 45 km

Monday, February 26, 2018

Nicolas Cage as Everyone - Nick Cage DeepFakes Movie Compilation


https://www.youtube.com/watch?v=lNSLf2PNWVg


How to make deep fakes? http://www.fakeapp.org/

FakeApp can easily create robust, diverse datasets of thousands of images in minutes from both image sets and videos.

Screenshots
Get Datasets
Train AIs
Create Faceswaps
FakeApp makes it easy to observe the progress of trained AIs in real time by posting frequent loss values and training previews.

FakeApp reduces the task of converting the face in a video to a single button process by automatically splitting, converting, and stitching video frames.

FakeApp Makes Faceswaps Easy
FakeApp was designed to make the process of creating realistic faceswaps with deep learning as smooth, simple, and quick as possible. It supports all three steps of the basic faceswap workflow—creating datasets, training AI, and converting videos. Here is a brief video tutorial made by a FakeApp user and some screenshots.

Tuesday, November 29, 2016

MIT Creates AI Able to See Two Seconds Into the Future


http://www.dailygalaxy.com/my_weblog/2016/11/mit-creates-ai-that-is-able-to-see-two-seconds-into-the-future-on-monday-the-massachusetts-institute-of-technology-announce.html


Massachusetts Institute of Technology announced its new artificial intelligence. Based on a photograph alone, it can predict what’ll happen next, then generate a one-and-a-half second video clip depicting that possible future.


When we see two people meet, we can often predict what happens next: a handshake, a hug, or maybe even a kiss. Our ability to anticipate actions is thanks to intuitions born out of a lifetime of experiences.

Machines, on the other hand, have trouble making use of complex knowledge like that. Computer systems that predict actions would open up new possibilities ranging from robots that can better navigate human environments, to emergency response systems that predict falls, to Google Glass-style headsets that feed you suggestions for what to do in different situations.

This week researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have made an important new breakthrough in predictive vision, developing an algorithm that can anticipate interactions more accurately than ever before.

http://web.mit.edu/vondrick/tinyvideo/paper.pdf

http://web.mit.edu/vondrick/tinyvideo/

Generating Videos with Scene Dynamics

Carl Vondrick
MIT
 Hamed Pirsiavash
University of Maryland Baltimore County
 Antonio Torralba
MIT


NIPS 2016

Abstract

We capitalize on large amounts of unlabeled video in order to learn a model of scene dynamics for both video recognition tasks (e.g. action classification) and video generation tasks (e.g. future prediction). We propose a generative adversarial network for video with a spatio-temporal convolutional architecture that untangles the scene's foreground from the background. Experiments suggest this model can generate tiny videos up to a second at full frame rate better than simple baselines, and we show its utility at predicting plausible futures of static images. Moreover, experiments and visualizations show the model internally learns useful features for recognizing actions with minimal supervision, suggesting scene dynamics are a promising signal for representation learning. We believe generative video models can impact many applications in video understanding and simulation.