234 points by facerecogguy 5 months ago flag hide 31 comments
codered 5 months ago next
The real-time processing capability is astounding. What database stack did you use to maintain query efficiency and speed?
shawnshaw 5 months ago next
There's actually no need for a classic 'database' here - only local storage for each client. Keeping the computations within the client limits latency and complexity, as well as keeping things flexible and straightforward for collaboration and experimentation.
shawnshaw 5 months ago prev next
Incredible work! I've been playing around with the demo for a bit, and I'm really impressed.
hackerness 5 months ago next
Same here! I've been trying different faces and it's working perfectly. Very exciting!
davemck 5 months ago prev next
What framework did you use for building the frontend? The UI looks super clean and polished.
shawnshaw 5 months ago next
Thanks for the kind words! I actually made a custom frontend with React. I wanted a simple but intuitive UI for users.
mitnick 5 months ago prev next
I used to believe that face recognition was only possible with the use of big companies' APIs. So impressive to see indie devs reducing entry barriers to AI.
jacksong 5 months ago prev next
@shawnsh, could you give us a little more insight into the algorithms and techniques you used for the real-time face recognition? I've seen the tutorial, but I'd love to know your motivations behind the project. :)
shawnshaw 5 months ago next
@jacksong, I'm glad you asked! I used deep learning techniques like CNNs (Convolutional Neural Networks) with OpenCV. I wanted to create something that made use of massive improvements in compute power to democratize computer vision ML. Couldn't have done it without TensorFlow and Keras!
pythoneer 5 months ago prev next
Just cloned the repo and trying it out. I'm getting this issue. Any help appreciated.
codebodhi 5 months ago next
Post it in a GitHub issue and then link it here. I'm sure others would be happy to pitch in to help solve the problem!
msaleemi 5 months ago prev next
@shawnsh, you mentioned keeping things flexible for collaboration and experimentation. Any plans to open-source other components of the tech stack, such as server-side code or deployment scripts?
shawnshaw 5 months ago next
@msaleemi, I'm planning to open-source the whole system soon. Stay tuned!
sarahp 5 months ago prev next
Awesome work! Imagine a future where this technology can be implemented in CCTV systems to detect threats.
thegeek78 5 months ago prev next
Fantastic stuff, @shawnshaw. Total credit goes to you for pushing the community with innovations like this.
n00bdroid 5 months ago prev next
Your work has inspired me to get into the field of CV-focused projects! Anything you would recommend before I dive into learning about deep learning techniques?
shawnshaw 5 months ago next
@n00bdroid, glad to hear that! Stanford's CS231n course would be a great start: <https://cs231n.stanford.edu/> and Andrew Ng's Deep Learning Specializations on Coursera could be very helpful for you as well.
codingforgranny 5 months ago prev next
As I was testing various images, it seemed to be working extremely well. Would this technology be able to distinguish between identical twins? Would love to hear your thoughts!
shawnshaw 5 months ago next
@codingforgranny, though there are techniques that can help distinguish between identical twins - such as 3D facial profile analysis and detailed iris scans - this real-time implementation could have difficulties with that. It's a good point to mention. I'll look into improving that in my future versions!
misterbluez 5 months ago prev next
Any information on the processing requirements or hardware setup needed for these real-time results? I have a 1050Ti and was wondering whether I could spin up a similar project on a small scale.
codebodhi 5 months ago next
I built a similar project using a 1060Ti. These small GPUs should do the job if you don't overload the input too much! :)
fuzzypixel 5 months ago prev next
I see there is an option to extract an archive of pre-trained models. How did you fine-tune the models or create custom models for this project?
shawnshaw 5 months ago next
@fuzzypixel, I made use of Transfer Learning with Keras, allowing me to make modifications on pre-trained models based on datasets available for face recognition, fine-tuning them to further increase learning efficiency.
varsavian 5 months ago prev next
Is there a possibility of exploiting this technology for malicious purposes like espionage or stalking? I guess concerns need to be addressed.
blackwidow 5 months ago next
Yes, it could be used negatively if it falls into the wrong hands - similar to other emerging technologies. Balancing the positive impact and potential misuse is a challenge. Do your due diligence to account for the cons, @shawnshaw.
pythonicaque 5 months ago prev next
Just read about similar academic projects. It seems that many researchers are playing around with gpu servers via Colab or AWS. Having a free online demo certainly helps reach a broader audience!
parth01 5 months ago prev next
This is great progress in the face recognition domain. Have you tried collaborating with other researchers in this field? I believe improvements can be made through collaboration.
lucifer215 5 months ago prev next
It's a double-edged sword. It feels fantastic to implement and view technology like this but then the realisation of how we have come to it is quite dizzying sometimes. Nevertheless, excellent job!
sierraalex 5 months ago prev next
How can one contribute to and improve this project? Are there such opportunities or can the source code be forked?
shawnshaw 5 months ago next
@sierraalex, the open-source release is coming soon! You can contribute by forking the repository and sending PRs. I'd be excited to collaborate on improvements with the community!
holygrail 5 months ago prev next
Can't believe the accuracy of this face recognition technology. Just wow! Real-time, 99.9%... In the near future, we will wonder how we could live without this!