There are a few days left in the new year. How are things going for you lately?
Express the reply by a reaction...
Express the reply by a reaction...
💯5💔2🍾2🥰1😢1😭1
Shitij's Happy Place.
There are a few days left in the new year. How are things going for you lately? Express the reply by a reaction...
Also, it's gonna be alright for those of you who did the sad/crying emoji. It's gonna come around. Has come for me so I can TELL. It will be alright.
🥰2👍1
Ladies and Gentlemen, Happy New Year ❤️. Hoping this year brings love, peace and happiness in our life. Cheers 🥂 to whatever situation we have been through and going through. After all it's the same time we all are living but just different events.
Once again, cheers to the new year! 🥂
Once again, cheers to the new year! 🥂
❤6🎉4🤡1💋1
Shitij's Happy Place.
Ladies and Gentlemen, Happy New Year ❤️. Hoping this year brings love, peace and happiness in our life. Cheers 🥂 to whatever situation we have been through and going through. After all it's the same time we all are living but just different events. Once again…
Give reaction to this post lads.
❤2
The sudden urge to just burn everything of mine, including me, the person who has built so many connections with people, built projects with guys I thought I could never have in my life, the ideologies. Everything. I wish I could turn off my brain for sometime and just disappear from the face of the people I know. Feels so heavy yet it makes me so vulnerable.
👀3❤1
Big day.
The research paper I've been building towards for the past year just dropped as a preprint with a DOI. Submitted to a Springer journal, currently in peer review.
https://dx.doi.org/10.21203/rs.3.rs-9552341/v1
The project is GlyphMotion — a real-time 4K multi-object tracking pipeline. You feed it video, it tracks every object across every frame, and outputs annotated 4K with audio intact. The interesting engineering is in how it gets there.
Synchronous pipelines can't handle this workload. We built a three-thread async architecture — reader, inference, writer — with bounded queues between them. Jitter went from 337ms down to 27ms on modern hardware, and from 21 seconds down to 46ms on older hardware.
The finding that caught us off guard: raw 4K footage is actually bad for tracking. The sensor noise in uncompressed footage constantly throws off the tracker's identity associations. Compressing at CRF 24 before running inference stabilised it — MOTA jumped from 55 to 85. Compression improving accuracy is not intuitive, but the data is consistent across 159 videos.
The quality loss from compression is recovered through a layer we call HFDR — High-Frequency Detail Reinjection — which adds back the spatial detail that compression strips. Final average VMAF across 159 videos: 96.67.
Built with Sayan Sarkar @sayann70. Supervised by Dr. Kretika Goel, DIT University. We started this right after class 12 boards, before college.
Read the preprint if you're curious. Happy to answer questions.
The research paper I've been building towards for the past year just dropped as a preprint with a DOI. Submitted to a Springer journal, currently in peer review.
https://dx.doi.org/10.21203/rs.3.rs-9552341/v1
The project is GlyphMotion — a real-time 4K multi-object tracking pipeline. You feed it video, it tracks every object across every frame, and outputs annotated 4K with audio intact. The interesting engineering is in how it gets there.
Synchronous pipelines can't handle this workload. We built a three-thread async architecture — reader, inference, writer — with bounded queues between them. Jitter went from 337ms down to 27ms on modern hardware, and from 21 seconds down to 46ms on older hardware.
The finding that caught us off guard: raw 4K footage is actually bad for tracking. The sensor noise in uncompressed footage constantly throws off the tracker's identity associations. Compressing at CRF 24 before running inference stabilised it — MOTA jumped from 55 to 85. Compression improving accuracy is not intuitive, but the data is consistent across 159 videos.
The quality loss from compression is recovered through a layer we call HFDR — High-Frequency Detail Reinjection — which adds back the spatial detail that compression strips. Final average VMAF across 159 videos: 96.67.
Built with Sayan Sarkar @sayann70. Supervised by Dr. Kretika Goel, DIT University. We started this right after class 12 boards, before college.
Read the preprint if you're curious. Happy to answer questions.
❤4
Shitij's Happy Place. pinned «Big day. The research paper I've been building towards for the past year just dropped as a preprint with a DOI. Submitted to a Springer journal, currently in peer review. https://dx.doi.org/10.21203/rs.3.rs-9552341/v1 The project is GlyphMotion — a real…»
Also, yes you can check this project here @ProjectGlyphmotionStudiosBot
And here primarily at https://projectglyphmotion.studio/
You can give it a video which has objects like cats, dogs, humans, cars, boats, and other common objects.
It will track it frame by frame and return you to processed videos.
There are no limits for the number of videos that you process but there is a frame limit of 200000 frames.
To give an idea it's around 55 minutes of 60fps footage and 1 hours 51 minutes of 30fps footage.
YouTube links are accepted as well, and also any internet video link which is public, and also local uploads.
It can only process one video at a time, but there is a queue system which sets your video in a queue. First come first serve basis.
Why use this? Honestly, there is nothing for the users but for me which is the dataset, what dataset? When you process any video, a benchmark is triggered after its finished processing, that benchmark gives me the data of video quality metrics and other research data that I heavily need for my further work.
Definitely check this out and let me know what thoughts you can frame while using this project.
And here primarily at https://projectglyphmotion.studio/
You can give it a video which has objects like cats, dogs, humans, cars, boats, and other common objects.
It will track it frame by frame and return you to processed videos.
There are no limits for the number of videos that you process but there is a frame limit of 200000 frames.
To give an idea it's around 55 minutes of 60fps footage and 1 hours 51 minutes of 30fps footage.
YouTube links are accepted as well, and also any internet video link which is public, and also local uploads.
It can only process one video at a time, but there is a queue system which sets your video in a queue. First come first serve basis.
Why use this? Honestly, there is nothing for the users but for me which is the dataset, what dataset? When you process any video, a benchmark is triggered after its finished processing, that benchmark gives me the data of video quality metrics and other research data that I heavily need for my further work.
Definitely check this out and let me know what thoughts you can frame while using this project.
❤3