harpreet's banner

harpreet

@DataScienceHarp • 7,565 subscribers

Hacker-in-residence. open source contributor. shipping all the integrations. cvpr is better than neurips

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

anon account on reddit lost a handmade RC plane in the desert. 3,000 drone images. one carbon-fiber airframe. one shadow that looks exactly like it i built a detector that scores every dark blob on shape and surface texture: a plane is elongated, smooth, solid, and convex. packaged it as a fiftyone plugin so anyone can run it. collapsed 3,000 images into a ranked shortlist you can review in minutes. one image has a shadow that looks exactly like the plane. the pilot already drove out to check it in person. it was a decoy the dataset, the plugin, and the shortlist are all public start here, read the dataset card:

anon account on reddit lost a handmade RC plane in the desert. 3,000 drone images. one carbon-fiber airframe. one shadow that looks exactly like it i built a detector that scores every dark blob on shape and surface texture: a plane is elongated, smooth, solid, and convex. packaged it as a fiftyone plugin so anyone can run it. collapsed 3,000 images into a ranked shortlist you can review in minutes. one image has a shadow that looks exactly like the plane. the pilot already drove out to check it in person. it was a decoy the dataset, the plugin, and the shortlist are all public start here, read the dataset card:

63,322 views • 3 months ago

a robot can recognize every object in a room and still not know how to open a single drawer knowing there's a cabinet is scene understanding. knowing which handle to grab, which direction to pull, and that it's a drawer not a door is affordance understanding almost no dataset has it in 3D SceneFun3D: 14,800+ functional elements across 710 laser-scanned indoor scenes. every handle, knob, button, and switch labeled with its affordance, motion axis, motion type, and a natural language task description each scene's 3D point cloud with its iPad video recordings in fiftyone. the affordance annotations live in 3D and project into the video frames check it out here

a robot can recognize every object in a room and still not know how to open a single drawer knowing there's a cabinet is scene understanding. knowing which handle to grab, which direction to pull, and that it's a drawer not a door is affordance understanding almost no dataset has it in 3D SceneFun3D: 14,800+ functional elements across 710 laser-scanned indoor scenes. every handle, knob, button, and switch labeled with its affordance, motion axis, motion type, and a natural language task description each scene's 3D point cloud with its iPad video recordings in fiftyone. the affordance annotations live in 3D and project into the video frames check it out here

18,800 views • 3 months ago

i fine-tuned SAM3 on 1,800 aerial drone images. the loss went down. then it started labeling cars as boats, tricycles, and people all at once, on the same object two things were wrong and neither was obvious 1. i was training the entire vision encoder on 1,800 samples freezing the encoder and only training the mask decoder fixed the hallucinations immediately 2. my training data had zero negative examples, every prompt always matched a mask adding 30% negative pairs where wrong prompts map to empty masks fixed the rest two decisions i made so early i forgot they were decisions

i fine-tuned SAM3 on 1,800 aerial drone images. the loss went down. then it started labeling cars as boats, tricycles, and people all at once, on the same object two things were wrong and neither was obvious 1. i was training the entire vision encoder on 1,800 samples freezing the encoder and only training the mask decoder fixed the hallucinations immediately 2. my training data had zero negative examples, every prompt always matched a mask adding 30% negative pairs where wrong prompts map to empty masks fixed the rest two decisions i made so early i forgot they were decisions

10,136 views • 3 months ago

DeciLM-7B is here and it beats Mistral-7B on all benchmarks - topping the HF Open LLM Leaderboard in it's weight class! Here's what you need to hack around: 📔DeciLM-7B Base Model Notebook: 📘DeciLM-7B Fine-tuning Notebook: 📗DeciLM-7-Instruct Notebook: Help us get trending on HuggingFace by ❤️ the model cards 🤗 Base model: 🤗 Instruction-tuned model: 🤗 Demo on HF Spaces: 👀Check out DeciLM + Infery: 📑 Read more about the model in the release blog:

DeciLM-7B is here and it beats Mistral-7B on all benchmarks - topping the HF Open LLM Leaderboard in it's weight class! Here's what you need to hack around: 📔DeciLM-7B Base Model Notebook: 📘DeciLM-7B Fine-tuning Notebook: 📗DeciLM-7-Instruct Notebook: Help us get trending on HuggingFace by ❤️ the model cards 🤗 Base model: 🤗 Instruction-tuned model: 🤗 Demo on HF Spaces: 👀Check out DeciLM + Infery: 📑 Read more about the model in the release blog:

13,965 views • 2 years ago

No more content to load