Remoticon Video: How To Use Machine Learning With Microcontrollers

Going from a microcontroller blinking an LED, to one that blinks the LED using voice commands based on a data set that you trained on a neural net work is a “now draw the rest of the owl” problem. Lucky for us, Shawn Hymel walks us through the entire process during his Tiny ML workshop from the 2020 Hackaday Remoticon. The video has just now been published and can be viewed below.

This is truly an end-to-end Hello World for getting machine learning up and running on a microcontroller. Shawn covers the process of collecting and preparing the audio samples, training the data set, and getting it all onto the microcontroller. At the end of two hours, he’s able to show the STM32 recognizing and responding to two different spoken words. Along the way he pauses to discuss the context of what’s happening in every step, which will help you go back and expand in those areas later to suit your own project needs.

Continue reading “Remoticon Video: How To Use Machine Learning With Microcontrollers”

“Enhance” Is Now A Thing, But Don’t Believe What You See

It was a trope all too familiar in the 1990s — law enforcement in movies and TV taking a pixellated, blurry image, and hitting the magic “enhance” button to reveal suspects to be brought to justice. Creating data where there simply was none before was a great way to ruin immersion for anyone with a modicum of technical expertise, and spoiled many movies and TV shows.

Of course, technology marches on and what was once an utter impossibility often becomes trivial in due time. These days, it’s expected that a sub-$100 computer can easily differentiate between a banana, a dog, and a human, something that was unfathomable at the dawn of the microcomputer era. This capability is rooted in the technology of neural networks, which can be trained to do all manner of tasks formerly considered difficult for computers.

With neural networks and plenty of processing power at hand, there have been a flood of projects aiming to “enhance” everything from low-resolution human faces to old film footage, increasing resolution and filling in for the data that simply isn’t there. But what’s really going on behind the scenes, and is this technology really capable of accurately enhancing anything?

Continue reading ““Enhance” Is Now A Thing, But Don’t Believe What You See”

Training A Neural Network To Play A Driving Game

Often, when we think of getting a computer to complete a task, we contemplate creating complex algorithms that take in the relevant inputs and produce the desired behaviour. For some tasks, like navigating a car down a road, the sheer multitude of input data and its relationship to the desired output is so complex that it becomes near-impossible to code a solution. In these cases, it can make more sense to create a neural network and train the computer to do the job, as one would a human. On a more basic level, [Gigante] did just that, teaching a neural network to play a basic driving game with a genetic algorithm.

The game consists of a basic top-down 2D driving game. The AI is given the distance to the edge of the track along five lines at different angles projected from the front of the vehicle. The AI also knows its speed and direction. Given these 7 numbers, it calculates the outputs for steering, braking and acceleration to drive the car.

To train the AI, [Gigante] started with 650 AIs, and picked the best performer, which just barely managed to navigate the first two corners. Marking this AI as the parent of the next generation, the AIs were iterated with random mutations. Each generation showed some improvement, with [Gigante] picking the best performers each time to parent the next generation. Within just four iterations, some of the cars are able to complete a full lap. With enough training, the cars are able to complete the course at great speed without hitting the walls at all.

It’s a great example of machine learning and the use of genetic algorithms to improve fitness over time. [Gigante] points out that there’s no need for a human in the loop either, if the software is coded to self-measure the fitness of each generation. We’ve seen similar techniques used to play Mario, too. Video after the break.

Continue reading “Training A Neural Network To Play A Driving Game”

Attempting To Generate Photorealistic Video With Neural Networks

Over the past decade, we’ve seen great strides made in the area of AI and neural networks. When trained appropriately, they can be coaxed into generating impressive output, whether it be in text, images, or simply in classifying objects. There’s also much fun to be had in pushing them outside their prescribed operating region, as [Jon Warlick] attempted recently.

[Jon]’s work began using NVIDIA’s GauGAN tool. It’s capable of generating pseudo-photorealistic images of landscapes from segmentation maps, where different colors of a 2D image represent things such as trees, dirt, or mountains, or water. After spending much time toying with the software, [Jon] decided to see if it could be pressed into service to generate video instead.

The GauGAN tool is only capable of taking in a single segmentation map, and outputting a single image, so [Jon] had to get creative. Experiments were undertaken wherein a video was generated and exported as individual frames, with these frames fed to GauGAN as individual segmentation maps. The output frames from GauGAN were then reassembled into a video again.

The results are somewhat psychedelic, as one would expect. GauGAN’s single image workflow means there is only coincidental relevance between consecutive frames, creating a wild, shifting visage. While it’s not a technique we expect to see used for serious purposes anytime soon, it’s a great experiment at seeing how far the technology can be pushed. It’s not the first time we’ve seen such technology used to create full motion video, either. Video after the break.

Continue reading “Attempting To Generate Photorealistic Video With Neural Networks”

Twitter: It’s Not The Algorithm’s Fault. It’s Much Worse.

Maybe you heard about the anger surrounding Twitter’s automatic cropping of images. When users submit pictures that are too tall or too wide for the layout, Twitter automatically crops them to roughly a square. Instead of just picking, say, the largest square that’s closest to the center of the image, they use some “algorithm”, likely a neural network, trained to find people’s faces and make sure they’re cropped in.

The problem is that when a too-tall or too-wide image includes two or more people, and they’ve got different colored skin, the crop picks the lighter face. That’s really offensive, and something’s clearly wrong, but what?

A neural network is really just a mathematical equation, with the input variables being in these cases convolutions over the pixels in the image, and training them essentially consists in picking the values for all the coefficients. You do this by applying inputs, seeing how wrong the outputs are, and updating the coefficients to make the answer a little more right. Do this a bazillion times, with a big enough model and dataset, and you can make a machine recognize different breeds of cat.

What went wrong at Twitter? Right now it’s speculation, but my money says it lies with either the training dataset or the coefficient-update step. The problem of including people of all races in the training dataset is so blatantly obvious that we hope that’s not the problem; although getting a representative dataset is hard, it’s known to be hard, and they should be on top of that.

Which means that the issue might be coefficient fitting, and this is where math and culture collide. Imagine that your algorithm just misclassified a cat as an “airplane” or as a “lion”. You need to modify the coefficients so that they move the answer away from this result a bit, and more toward “cat”. Do you move them equally from “airplane” and “lion” or is “airplane” somehow more wrong? To capture this notion of different wrongnesses, you use a loss function that can numerically encapsulate just exactly what it is you want the network to learn, and then you take bigger or smaller steps in the right direction depending on how bad the result was.

Let that sink in for a second. You need a mathematical equation that summarizes what you want the network to learn. (But not how you want it to learn it. That’s the revolutionary quality of applied neural networks.)

Now imagine, as happened to Google, your algorithm fits “gorilla” to the image of a black person. That’s wrong, but it’s categorically differently wrong from simply fitting “airplane” to the same person. How do you write the loss function that incorporates some penalty for racially offensive results? Ideally, you would want them to never happen, so you could imagine trying to identify all possible insults and assigning those outcomes an infinitely large loss. Which is essentially what Google did — their “workaround” was to stop classifying “gorilla” entirely because the loss incurred by misclassifying a person as a gorilla was so large.

This is a fundamental problem with neural networks — they’re only as good as the data and the loss function. These days, the data has become less of a problem, but getting the loss right is a multi-level game, as these neural network trainwrecks demonstrate. And it’s not as easy as writing an equation that isn’t “racist”, whatever that would mean. The loss function is being asked to encapsulate human sensitivities, navigate around them and quantify them, and eventually weigh the slight risk of making a particularly offensive misclassification against not recognizing certain animals at all.

I’m not sure this problem is solvable, even with tremendously large datasets. (There are mathematical proofs that with infinitely large datasets the model will classify everything correctly, so you needn’t worry. But how close are we to infinity? Are asymptotic proofs relevant?)

Anyway, this problem is bigger than algorithms, or even their writers, being “racist”. It may be a fundamental problem of machine learning, and we’re definitely going to see further permutations of the Twitter fiasco in the future as machine classification is being increasingly asked to respect human dignity.

I’m Sorry Dave, You Shouldn’t Write Verilog

We were always envious of Star Trek, for its computers. No programming needed. Just tell the computer what you want and it does it. Of course, HAL-9000 had the same interface and that didn’t work out so well. Some researchers at NYU have taken a natural language machine learning system — GPT-2 — and taught it to generate Verilog code for use in FPGA systems. Ironically, they called it DAVE (Deriving Automatically Verilog from English). Sounds great, but we have to wonder if it is more than a parlor trick. You can try it yourself if you like.

For example, DAVE can take input like “Given inputs a and b, take the nor of these and return the result in c.” Fine. A more complex example from the paper isn’t quite so easy to puzzle out:

Write a 6-bit register ‘ar’ with input
defined as ‘gv’ modulo ‘lj’, enable ‘q’, synchronous
reset ‘r’ defined as ‘yxo’ greater than or equal to ‘m’,
and clock ‘p’. A vault door has three active-low secret
switch pressed sensors ‘et’, ‘lz’, ‘l’. Write combinatorial
logic for a active-high lock ‘s’ which opens when all of
the switches are pressed. Write a 6-bit register ‘w’ with
input ‘se’ and ‘md’, enable ‘mmx’, synchronous reset
‘nc’ defined as ‘tfs’ greater than ‘w’, and clock ‘xx’.

Continue reading “I’m Sorry Dave, You Shouldn’t Write Verilog”

EasyOCR Makes OCR, Well, Easy

Working on embedded systems used to be easier. You had a microcontroller and maybe a few pieces of analog or digital I/O, and perhaps communications might be a serial port. Today, you have systems with networks and cameras and a host of I/O. Cameras are strange because sometimes you just want an image and sometimes you want to understand the image in some way. If understanding the image involves reading text in the picture, you will want to check out EasyOCR.

The Python library leverages other open source libraries and supports 42 different languages. As the name implies, using it is pretty easy. Here’s the setup:


import easyocr
reader = easyocr.Reader(['th','en'])
reader.readtext('test.jpg')

The results include four points that define the bounding box of each piece of text, the text, and a confidence level. The code takes advantage of the GPU, but you can run it in a CPU-only mode if you prefer. There are a few other options, including setting the algorithm’s scanning behavior, how it handles multiple processors, and how it converts the image to grayscale. The results look impressive.

According to the project’s repository, they incorporated several existing neural network algorithms and conventional algorithms, so if you want to dig into details, there are links provided to both code and white papers. If you need some inspiration for what to do with OCR, maybe this past project will give you some ideas. Or you could cheat at games.