Perfected state of the art "object detection"

fingers crossed not a repost 😣

FML. All those "train your own object detection" articles can go burn in hell. Not even TFs pertained models work!

(Warning: kinda long && somewhat of a political rant)

Every time I tell someone I work with AI, the first thing to come out of their mouth is "oh but AI is going to take over the world!"

No.

It was only somewhat recently that it started being able to recognize what was in a picture from over 3 million images, and that too it's not that great at. Honestly people always say "AI is just if-else" ironically, but it isn't really that far from the truth, we just multiply an input by weights and check the output.

It isn't some magical sauce, it's not being born and then exploring a problem, it's just glorified-probability prediction. Even in "unsupervised" learning, the domain set is provided; in "reinforcement learning" which has gotten super popular lately we just have the computer decide which policy is optimal and apply that to an environment. It's a glorified decision tree (and technically tree models like XGBoost outperform neural networks and deep learning on a large number of problems) and it isn't going to "decide" to take over the planet.

Honestly all of this is just born out of Elon Musk fans who take his word as truth and have been led to believe that AI is going to take over the world. There are a billion reasons why it can't! And to top it off this takes away a lot of public attention from VERY concerning ethical issues with AI.

Am I the only one who saw Google Duplex being unveiled and immediately thought "fraud"? Forget phone scammers, if you trained duplex on the mannerisms of, for example, a famous politician's voice, you could impersonate them in an audio clip (or even video clip with deepfakes). Or for example the widespread use of object detection and facial recognition in surveillance systems deployed by DoD. Or the use of AI combined with location tracking and browsing analytics for targeted marketing.

The list of ethics breaches are endless, and I find it super suspicious that those profiting the most off of unethical AI are all too eager to shift public concern to some science fiction Terminator style takeover that, if ever possible, would be a long way out and is not any sort of a priority issue right now.

Did I ever tell you kids about the time I worked for a company that got a contract to develop an iOS application around some object detection software that had been developed by another team?

Company I was working for was a tiny software consultancy, and this was my first ever dev job (I’m at my second now 😅). Nobody at the company has experience building mobile applications but CEO decides that the app should be written in React Native because _he_ knows React Native.

During a meeting with the client, CEO jokes about how easy the ask is and says he could finish it in a weekend. Please note that Head of Engineering had already budgeted a quarter for the work. CEO says we can do it in a week! And moves up the deadline. And only assigns two engineers to project. I am not one of those engineers.

The two engineers that are put on it struggle. A lot. They can’t seem to get the object detection to work at all, and the code that’s already written is in Objective-C. I realize one of the issues is that the engineers on the project can’t read Objective-C because they have no experience with Objective-C or even C. I have experience with C, so I volunteer to take a look at it to try to see what’s going on.

Turns out the problem is that the models are trained on one type of image format and the iPhone camera takes images in a different format.

The end of the week comes, they do not succeed in figuring out the image conversion in React Native. There’s an in-person demo with the customers scheduled for the next Monday. CEO spends the weekend trying to build the app. Only succeeds in locking literally every other engineer out of the project.

They manage to negotiate a second chance where we deliver what we were supposed to deliver at the original schedule.

I spent the weekend looking up how to convert images and figure it would be a lot easier to interface with the Objective-C if we used Swift. Taught myself enough Swift over the weekend to feel dangerous. Spoke to Head of Engineering on Monday and proposed solution — start over in Swift. Volunteer to lead effort. Eventually convince them it’s a good idea (and really, what’s the worst that can happen? If this solves our main problem at the moment, that’s still more progress than the original team made)

Spend the next week working 16 hour days building out application. Meet requirements for next deadline. Save contract.

And that’s ONE of the stories of my first dev job that got me hired as a senior engineer despite only having 10 months of work experience in the industry.

python machine learning tutorials:
- import preprocessed dataset in perfect format specially crafted to match the model instead of reading from file like an actual real life would work
- use images data for recurrent neural network and see no problem
- use Conv1D for 2d input data like images
- use two letter variable names that only tutorial creator knows what they mean.
- do 10 data transformation in 1 line with no explanation of what is going on
- just enter these magic words
- okey guys thanks for watching make sure to hit that subscribe button

ehh, the machine learning ecosystem is burning pile of shit let me give you some examples:

- thanks to years of object oriented programming research and most wonderful abstractions we have "loss.backward()" which have no apparent connection to model but it affects the model, good to know
- cannot install the python packages because python must be >= 3.9 and at the same time < 3.9
- runtime error with bullshit cryptic message
- python having no data types but pytorch forces you to specify float32
- lets throw away the module name of a function with these simple tricks:
"import torch.nn.functional as F"
"import torch_geometric.transforms as T"
- tensor.detach().cpu().numpy() ???
- class NeuralNetwork(torch.nn.Module):
def __init__(self):
super(NeuralNetwork, self).__init__() ????
- lets call a function that switches on the tracking of math operations on tensors "model.train()" instead of something more indicative of the function actual effect like "model.set_mode_to_train()"
- what the fuck is ".iloc" ?
- solving environment -/- brings back memories when you could make a breakfast while the computer was turning on
- hey lets choose the slowest, most sloppy and inconsistent language ever created for high performance computing task called "data sCieNcE". but.. but. you can use numpy! I DONT GIVE A SHIT about numpy why don't you motherfuckers create a language that is inherently performant instead of calling some convoluted c++ library that requires 10s of dependencies? Why don't you create a package management system that works without me having to try random bullshit for 3 hours???
- lets set as industry standard a jupyter notebook which is not git compatible and have either 2 second latency of tab completion, no tab completion, no documentation on hover or useless documentation on hover, no way to easily redo the changes, no autosave, no error highlighting and possibility to use variable defined in a cell below in the cell above it
- lets use inconsistent variable names like "read_csv" and "isfile"
- lets pass a boolean variable as a string "true"
- lets contribute to tech enabled authoritarianism and create a face recognition and object detection models that china uses to destroy uyghur minority
- lets create a license plate computer vision system that will help government surveillance everyone, guys what a great idea

I don't want to deal with this bullshit language, bullshit ecosystem and bullshit unethical tech anymore.

Annoying Indian professors are everywhere. It's a computer vision class are you really teaching us Regression?

What about transfer learning? object detection! Give us papers to read, let's do projects.. what the hell is this I am going take attendance bullshit and teaching crappy concepts.

I did not sign up for this shit! I came here for my Masters to get away from pompous mother fuckers like you ...

My class is also filled with those idiots,who think bias in a neural network is somehow related to class imbalance ? Now the same idiot proceeds to ask questions like...

Why would the weights change in a neural network?
Motherfucker why you in this class ? Why don't you stick to your shit and ask these questions later..

I am so pissed off right now guys ...
I was sitting in my lab understanding the deeper insights of BN, activation fucntions.. various optimizers ..etc Stuff that this idiot motherfucking teacher must be covering... UGGH.

I shouldn't cuss so much.. or at least add variety to my cuss words..

I am pissed off cuz instead of learning the shit I should be learning I am forced to come and attend this class and waste 2 hrs of my life ...

It's the summer i find it hard to focus anyway (want to go out hiking or swimming or something.) BUT. the moment I find some resolve to focus

I get this fucking bullshit.. !

My mind is so fucked right now... I can't think of anything but standing up in class and screaming " Mother fucker, mother fucker...(point to the idiots in class you) motherfuckers shut the fuck up..

Can someone suggest some colorful swear words ?

My brains not working -_-

It is just about now that I start feeling like "Anger" from inside out

Designed a person detection and tracking algorithm based on RCNN and lukas-kanade object tracking algorithm in openCV python.

Need help in cases of occlusion any suggestions?

Hey!
Has anyone figured out a way to easily run object detection on the raspberry pi3?
I'm having a really bad fps will running it.
Is there any optimised way to run?

Thanks in advance :')

I'm new in deep learning world, I want to develop a project regarding live text detection on mobile camera (without taking an image by the camera), can I do it by the tesorflow object detection api ? or with anyother tensorflow api or with opencv, which way is best? other suggestions are also welcome.

I’ve been researching the different YOLO object detection frameworks and I saw that YOLOv8 came out recently. It’s seems like the past couple versions (v5, v6, v7, and v8) have come out super fast. Anyone else notice this?