harpreet
@DataScienceHarp
Followers
7K
Following
27K
Media
967
Statuses
6K
🤖 👨🏽💻 Hacker-in-residence @voxel51| ❤️open source deep learning | VLMs| Visual AI| Learn. Hack. Write. Teach. Repeat. 🪯
I ship daily
Joined April 2020
Labeling data has got to be the most time consuming task I’ve ever done in deep learning.
2
0
4
Register here:. Session 1: Session 2: Session 3:
voxel51.com
Welcome to the Visual Agents Workshop Series, your virtual pass to some of the groundbreaking research, and innovations that defined this year’s conference.
0
0
1
I'm teaching a series of workshops in August that go deep into Visual Agents (specifically GUI Agents). The last session is fine-tuning a VLM for this task. A huge challenge is building a dataset for fine-tuning. I want annotations in COCO Format, but nothing exists for this,
4
0
2
Older. 3.1 on a 386.
0
0
4
Will have to integrate this into FiftyOne asap!.
all modality RAG 🔥. ColQwen-Omni is a new multimodal retrieval model that can retrieve anything (videos, audios, documents and more!). use with transformers 🤗.here's a smol demo on video retrieval ↙️
1
0
3
Had an awesome time visiting @huggingface office in Paris! Thank you @mervenoyann for the invite. Good you finally meet you and the legend @reach_vb in person. Looking forward to the next time. Cheers!
1
2
44
Does no one remember what happened with Inflection?.
I’m sorry but an executive founder leaving their company seems like an obvious fucking breach of fiduciary duty and massive fucking conflict of interest.
1
0
0
RT @Voxel51: The era of single-purpose vision models is ending. Introducing agglomerative vision models: a fundamental shift from how we’ve….
0
2
0
RT @Voxel51: If you're in Germany, come out and join us for a hands-on computer vision workshop on Advanced Car Damage Detection with Fifty….
0
1
0
Resources:. ⭐️ the repo on GitHub: 👨🏽💻 Notebook to get started:
github.com
Integrating OS-Atlas Base into FiftyOne as a Remote Source Zoo Model - harpreetsahota204/os_atlas
0
0
3
OS Atlas 7B is a solid vision model that will localize UI elements reliably, even when you deviate from their suggested prompts. Here's what I learned after two days of experimentation ⇣. 1) OS Atlas 7B reliably localizes UI elements even with prompt variations. • The model
1
1
5
Notebooks:. Natural images: OCR: Star the repo on GitHub:
github.com
Implementing Llama-3.1-Nemotron-Nano-VL-8B-V1 as a Remote Zoo Model for FiftyOne - harpreetsahota204/Nemotron_Nano_VL
0
0
3
Two days with Nemotron Nano VL taught me that it can spot a left leg in a crowd but can't find a button on a screen. It's surprisingly capable at natural images but completely breaks on UI tasks. Here are my main takeaways. 1. It's surprisingly good at natural images, despite
1
0
3
RT @Voxel51: 🎙️ Speakers include:. - @tuanacelik, @llama_index.- Julien Simon, @arcee_ai.- @NielsRogge, @huggingface.- Gabriel Trégoat, @Pr….
0
1
0
Very nice! You could also use FifityOne for this (and more) at a larger scale. Both Qwen2.5 VL are integrated as well:. • •
github.com
Moondream2 implementation as a remotely sourced zoo model for FiftyOne - harpreetsahota204/moondream2
Updated my HF Space for vibe testing smol VLMs on object detection, visual grounding, keypoint detection & counting! 👓. 🆕Compare Qwen2.5 VL 3B vs Moondream 2B side-by-side with annotated images & text outputs. Try examples or test your own images! 🏃👇
0
2
3
Notebook for integration in FO: Star the repo:
github.com
Implementing UI-TARS-1.5 as a Remote Zoo Model for FiftyOne - harpreetsahota204/UI_TARS
0
0
2