DataScienceHarp's profile picture. 🤖 👨🏽‍💻 Hacker-in-residence @voxel51| ❤️open source deep learning | VLMs| Visual AI| Learn. Hack. Write. Teach. Repeat. 🪯

harpreet

@DataScienceHarp

🤖 👨🏽‍💻 Hacker-in-residence @voxel51| ❤️open source deep learning | VLMs| Visual AI| Learn. Hack. Write. Teach. Repeat. 🪯

i literally can't comprehend the hype for docling, it's slow af for a 258M parameter model that only does one thing


Well you told me about nowhere and it sounds like some place I’d like to go


harpreet reposted

🚗📷 How can Vision-Language Models transform computer vision in industry? At #ODSCWest, @DataScienceHarp (@voxel51) leads a workshop on Visual AI for vehicle damage detection using the CarDD dataset + FiftyOne. 📅 Oct 28–30 | SF 🔗 hubs.li/Q03GX0jy0 #VisualAI #VLMs

_odsc's tweet image. 🚗📷 How can Vision-Language Models transform computer vision in industry?

At #ODSCWest, @DataScienceHarp (@voxel51) leads a workshop on Visual AI for vehicle damage detection using the CarDD dataset + FiftyOne.

📅 Oct 28–30 | SF
🔗 hubs.li/Q03GX0jy0 

#VisualAI #VLMs

I saw FastVLM at CVPR Nashville and been patiently waiting for Transformers integration ever since. It's here now. • Every vision token adds LLM latency. ViT generates 576 tokens at 336×336. That's 576 forward passes through your decoder. • FastVLM uses 144 tokens at 768×768…


So pumped to be here now

Later today Oasis will play the first of two sold out concerts at Rogers Stadium in Toronto 🇨🇦 📹 Arlen Ekstein / lilbmxharo



harpreet reposted

Traditional annotation is slowing down the race to fully autonomous vehicles. On Sept. 4, join @Porsche, Voxel51, and @databricks for an exclusive look at how cutting-edge VLMs, automated data pipelines, and human-in-the-loop feedback loops are transforming the way AV systems…

Voxel51's tweet image. Traditional annotation is slowing down the race to fully autonomous vehicles. 

On Sept. 4, join @Porsche, Voxel51, and @databricks for an exclusive look at how cutting-edge VLMs, automated data pipelines, and human-in-the-loop feedback loops are transforming the way AV systems…

harpreet reposted

@nvidiaomniverse NuRec and @Voxel51 FiftyOne integration tackles a major bottleneck in AV development when working with multi-sensor datasets. As AV systems move from R&D to deployment, issues like misaligned sensor calibrations, drifting ego-poses, and timestamps often introduce…


harpreet reposted

In Part 1, @datascienceharp will walk through: 👉 Why standard vision models fail catastrophically on GUI tasks 👉 The annotation bottlenecks that make GUI datasets so expensive to create 👉 The platform fragmentation that makes "click a button" mean 20 different things across…


Loading...

Something went wrong.


Something went wrong.