There are probably a few kinks to work out that will come up in practice!
https://gist.github.com/YashasSamaga/e2b19a6807a13046e399f4b... (download links for yolov4.weights is at https://github.com/AlexeyAB/darknet)
Using this, you will be able to detect if a cat is present in your image.
I plan (if I ever do this) to program a decay over time, starting at 100% chance/zero seconds, and moving to lower chance and higher random time interval.
It always seemes assume that I'm trying to buy something and tries to find products me that are somehow visually related to what it's looking at, or else to the content of some lettering that it's able to detect.
So I gave up on it.
Are there some settings I can tweak?
If you need high accuracy, let's say for e.g. estimating eco system performance based on specific plant distribution, 75% is very low (especially if you want to feed it into another predictor) compared to a professional field biologist.
That's awesome they have the assignment for download.
Love the info site design.
I expect, as they allude to, it will begin with simple classification tasks in order to stick with the clean user experience they've built. But I'm super eager to see what they propose in this area.
There are other options for this too, I think spacy.io has an annotation app.
We specialize in tabular data and are building a pipeline-based approach for creating and serving models.
In fact, our process for every feature we work on is the same—we start thinking about it with the way users are going to learn about it in mind, that allows us to simplify the way we talk about it and massage the messaging as much as possible, so when we have to talk about it externally, it's so tested that it just comes natural to us, and hopefully to the world.
Edit: yep, on Chrome Mobile I actually see an animation and stuff seems to work. On Firefox it's borked.
Reading the license I assume it may change at some future version to require money to use it, and that a new version will install and then say please pay us to continue using? Or probably just this product is no longer available? Note these are not things I am thinking will happen but rather my theoretical assumptions to try to answer the question of why has Microsoft, a for profit company, made this closed source, free tool that I think might be pretty useful for a lot of people.
Lobe will always let you train custom machine learning for free on your computer. We hope this becomes a vibrant ecosystem, and the business model around the edges can come later for value-add services.
ah ok, fine, just the legalese was making me wonder. And of course that we are in a capitalist system so not sure I follow the value for Microsoft in this scenario, but I guess you find making machine learning more available to apps somehow drives value.
So thanks for what looks like a pretty nice tool.
You agree to receive these automatic updates without any additional notice. Updates may not include or support all existing software features, services, or peripheral devices.
This is the part that causes me to assume it will stop working at some point in the future? And when would that be:
a) Term.
The term of this agreement will continue until the commercial release of the software. We also may not release a commercial version.
But honestly unsure if I am just paranoid. Or even if paranoia is the right term for my feeling about it, it's something Microsoft is letting me use and at some point it won't be usable anymore - such is life - might be the more reasonable response to it.
However I tried to train it to recognize some images of characters from an anime (so a little different than facial recognition), and I managed to break the model: achieving 64% error with significant number of examples per class. I think one downside is Lobe doesn't expose how potentially overconfident the model is. I would love the ability to take the existing model and test it on a new image that I can import into the app.
EDIT: I would love to see the following in a future version:
1. What are the percentages associated with each image per class. I see that an image was misclassified, but did it at least include my desired class in its top 5 predicted classes?
2. Test the model on unlabeled inputs directly in the app to see how well the model might generalize. I would like to see a "Test" tab on the left once training is complete.
3. View other metrics of model goodness like F-1 score and training details like CV partitions in the app somehow.
Again, this is a really cool idea :)
Here's a few tips for now: 1. You can view by "Test Images" on the Train tab (view options). So you can see how well your model is performing on your test images (a random 20% split from all of your images). 2. You can test your model on the Play tab, by dragging in new images your model has not seen, to see how well it is performing. You can also tell Lobe if it was correct or not and iteratively improve your model.
Check settings -> export -> local Api
In some of the less harmless applications of computer vision and machine learning, sometimes it will have very severe consequences for real people that a computer says yes or no when it really doesn't have the information to say either or. Some people are afraid of what will happen to society when these systems become as accurate as humans - I am honestly more worried about what will happen if they don't.
While I never got approved for that beta (probably rightly so, I'm just some random person with no actual connection to ML or AI), I was excited to see what their work led to. Congrats on releasing this latest iteration and acquisition!
The reasoning behind the change and the why we abstracted some of those details you are mentioning was to actually make it even more accessible for people to be able to build machine learning models. We think that this is a paradigm that should be used by everyone, and that’s why stripping down the onion of complexity was really important for us when we started with this project.
1. Collecting & labeling images 2. Training your model and evaluating the results 3. Playing with your model and seeing how well its performing
More info on AutoML: https://cloud.google.com/automl
* Easy to use - no coding, cloud configuration or machine learning experience required.
* Free & private - train for free on your own computer without uploading your data to the cloud. No accounts required.
* Ship anywhere - available for both Mac and Windows. Export your model and ship it on any platform you choose.
AutoML requires paid accounts with high friction setup and is focused on just training a model on your data. You would have to pay and retrain your model manually every time you want to make an iteration. Lobe gives fluidity with iterating and providing feedback to your model through Play.
Both apps are great!
This is next to ridiculous. I don't need an app or any assistance in counting my reps. I can do that myself. That's easy.
What I really dream of an app for is app to tell my mistakes in technique/posture for every particular exercise. I don't even mind putting a funny costume or some motion sensors on to make its job easier.
On the other comment, yes! The app you are describing sounds really interesting, and it is something that could be build using image classification, you just need the right images and camera setup, though!
On the other - the way you do an exercise, small details in your posture and the sequence of changes in it - that's what decides if what you do is going to make you more fit/strong, have no effect or just cause pure harm. It's extremely important (at least, in the beginning) to have somebody qualified to watch how you do it and correct you. Many people prefer to train alone though so they need such an app.
While the UI is quite nice (thanks to Mike Matas, I am sure), I don't see a strong advantage to using this on MacOs, when CreateML is available. CreateML doesn't have the simple interface of Lobe, but the UI is quite accessible and gives you access to additional classifies, like sound, text and tabular data. If you need ever more power, you can use TuriCreate if you want to stay in the Apple ecosystem.
The simplicity of the UI is a feature, but also a disadvantage when you start having more than a handful of labels and training images. I totally see how Lobe could be a nice intro into the world of labelling and classification.
Would you mind to elaborate on this?
Do you use Outlook? -- If there is interest, I can try and resurrect it. Although it's not as necessary as it once was -- not as many "Re: re: FW: re: fw: hello!" messages now that people use Slack and Teams, etc.
However, they do have opensourced bootstrap apps here: https://github.com/lobe
There would be a lot of cool ways to improve the model by giving feedback, either showing training images where the model is uncertain, or some more advanced explanations for classifications flagged as incorrect, in order to guide the user to gather the training data that can improve it.
And possibly providing a summary of where it knows it works well.
There are a lot of benefits there, both for improving models people are building but also to help users understand why their model is performing as it does.
its awesome for noobs like me to train things.
Thanks.
You should release this on android and market it hard. Remember machine learning will flourish when idiots like me can train to do to mundane task.
Boo hoo, I'm running Linux on my desktop...
